Programming for Data Science - Group Assignment
Lim Wen Yan (S2019787), Samuel Wong Ing Shiing (S2018282)
1st January 2021
Telco Customer Churn and Customer Service Call Prediction
Introduction
The dataset used in this project is a dataset with 21 variables and 3758 observations on Telecommunication Customer Churn Analytics. This dataset and project objective is revolving around one major problem with telecommunication operators which is customer retention. Over the years, operators have accustomed themselves to the practise where the customers with higher likelihood of cancelling the company’s service are identified and given special promotion to as an attempt to prevent subscription cnacellation and increase customer retention.
With that, the ablity to effectively identify this class of customers who are relatively likely to bail on the company is crucial. Hence, in this dataset, the normal characteristics and normal day-to-day usage of each customer is detailed to hopefully be used to predict the likelihood of each of them cancelling the service of the telecom company.
Import Relevant Libraries
library(tidyverse)## ── Attaching packages ─────────────────────────────────────── tidyverse 1.3.0 ──## ✓ ggplot2 3.3.2 ✓ purrr 0.3.4
## ✓ tibble 3.0.4 ✓ dplyr 1.0.2
## ✓ tidyr 1.1.2 ✓ stringr 1.4.0
## ✓ readr 1.4.0 ✓ forcats 0.5.0## Warning: package 'ggplot2' was built under R version 3.6.2## Warning: package 'tibble' was built under R version 3.6.2## Warning: package 'tidyr' was built under R version 3.6.2## Warning: package 'readr' was built under R version 3.6.2## Warning: package 'purrr' was built under R version 3.6.2## Warning: package 'dplyr' was built under R version 3.6.2## ── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──
## x dplyr::filter() masks stats::filter()
## x dplyr::lag() masks stats::lag()library(gtools)## Warning: package 'gtools' was built under R version 3.6.2library(ggplot2)
library(caret)## Loading required package: lattice##
## Attaching package: 'caret'## The following object is masked from 'package:purrr':
##
## liftlibrary(corrplot)## corrplot 0.84 loadedlibrary(viridis)## Loading required package: viridisLitelibrary(hrbrthemes)## NOTE: Either Arial Narrow or Roboto Condensed fonts are required to use these themes.## Please use hrbrthemes::import_roboto_condensed() to install Roboto Condensed and## if Arial Narrow is not on your system, please see https://bit.ly/arialnarrow
Import, View and Bind dataset
train_data <- read.csv("churnTrain.csv")
test_data <- read.csv("churnTest.csv")View first 6 rows of train dataset
head(train_data)## State Account_Length Area_Code Phone_No International_Plan Voice_Mail_Plan
## 1 KS 128 415 3824657 no yes
## 2 OH 107 415 3717191 no yes
## 3 NJ 137 415 3581921 no no
## 4 OH 84 408 3759999 yes no
## 5 OK 75 415 3306626 yes no
## 6 AL 118 510 3918027 yes no
## No_Vmail_Messages Total_Day_minutes Total_Day_Calls Total_Day_charge
## 1 25 265.1 110 45.07
## 2 26 161.6 123 27.47
## 3 0 243.4 114 41.38
## 4 0 299.4 71 50.90
## 5 0 166.7 113 28.34
## 6 0 223.4 98 37.98
## Total_Eve_Minutes Total_Eve_Calls Total_Eve_Charge Total_Night_Minutes
## 1 197.4 99 16.78 244.7
## 2 195.5 103 16.62 254.4
## 3 121.2 110 10.30 162.6
## 4 61.9 88 5.26 196.9
## 5 148.3 122 12.61 186.9
## 6 220.6 101 18.75 203.9
## Total_Night_Calls Total_Night_Charge Total_Intl_Minutes Total_Intl_Calls
## 1 91 11.01 10.0 3
## 2 103 11.45 13.7 3
## 3 104 7.32 12.2 5
## 4 89 8.86 6.6 7
## 5 121 8.41 10.1 3
## 6 118 9.18 6.3 6
## Total_Intl_Charge No_CS_Calls Churn
## 1 2.70 1 FALSE
## 2 3.70 1 FALSE
## 3 3.29 0 FALSE
## 4 1.78 2 FALSE
## 5 2.73 3 FALSE
## 6 1.70 0 FALSEView first 6 rows of test dataset
head(test_data)## State Account_Length Area_Code Phone_No International_Plan Voice_Mail_Plan
## 1 HI 101 510 3548815 no no
## 2 MT 137 510 3817211 no no
## 3 OH 103 408 4119481 no yes
## 4 NM 99 415 4189100 no no
## 5 SC 108 415 4133643 no no
## 6 IA 117 415 3756180 no no
## No_Vmail_Messages Total_Day_minutes Total_Day_Calls Total_Day_charge
## 1 0 70.9 123 12.05
## 2 0 223.6 86 38.01
## 3 29 294.7 95 50.10
## 4 0 216.8 123 36.86
## 5 0 197.4 78 33.56
## 6 0 226.5 85 38.51
## Total_Eve_Minutes Total_Eve_Calls Total_Eve_Charge Total_Night_Minutes
## 1 211.9 73 18.01 236.0
## 2 244.8 139 20.81 94.2
## 3 237.3 105 20.17 300.3
## 4 126.4 88 10.74 220.6
## 5 124.0 101 10.54 204.5
## 6 141.6 68 12.04 223.0
## Total_Night_Calls Total_Night_Charge Total_Intl_Minutes Total_Intl_Calls
## 1 73 10.62 10.6 3
## 2 81 4.24 9.5 7
## 3 127 13.51 13.7 6
## 4 82 9.93 15.7 2
## 5 107 9.20 7.7 4
## 6 90 10.04 6.9 5
## Total_Intl_Charge No_CS_Calls
## 1 2.86 3
## 2 2.57 0
## 3 3.70 1
## 4 4.24 1
## 5 2.08 2
## 6 1.86 1Bind Test and Train dataset from Kaggle for easier data preprocessing
df <- smartbind(train_data, test_data)
View the Summary, Structure and Dimension of the dataset
# view the first 6 rows of the train_data
head(df)## State Account_Length Area_Code Phone_No International_Plan Voice_Mail_Plan
## 1:1 KS 128 415 3824657 no yes
## 1:2 OH 107 415 3717191 no yes
## 1:3 NJ 137 415 3581921 no no
## 1:4 OH 84 408 3759999 yes no
## 1:5 OK 75 415 3306626 yes no
## 1:6 AL 118 510 3918027 yes no
## No_Vmail_Messages Total_Day_minutes Total_Day_Calls Total_Day_charge
## 1:1 25 265.1 110 45.07
## 1:2 26 161.6 123 27.47
## 1:3 0 243.4 114 41.38
## 1:4 0 299.4 71 50.90
## 1:5 0 166.7 113 28.34
## 1:6 0 223.4 98 37.98
## Total_Eve_Minutes Total_Eve_Calls Total_Eve_Charge Total_Night_Minutes
## 1:1 197.4 99 16.78 244.7
## 1:2 195.5 103 16.62 254.4
## 1:3 121.2 110 10.30 162.6
## 1:4 61.9 88 5.26 196.9
## 1:5 148.3 122 12.61 186.9
## 1:6 220.6 101 18.75 203.9
## Total_Night_Calls Total_Night_Charge Total_Intl_Minutes Total_Intl_Calls
## 1:1 91 11.01 10.0 3
## 1:2 103 11.45 13.7 3
## 1:3 104 7.32 12.2 5
## 1:4 89 8.86 6.6 7
## 1:5 121 8.41 10.1 3
## 1:6 118 9.18 6.3 6
## Total_Intl_Charge No_CS_Calls Churn
## 1:1 2.70 1 FALSE
## 1:2 3.70 1 FALSE
## 1:3 3.29 0 FALSE
## 1:4 1.78 2 FALSE
## 1:5 2.73 3 FALSE
## 1:6 1.70 0 FALSE# view the dimension of the df
dim(df)## [1] 3758 21# view the structure of the df
str(df)## 'data.frame': 3758 obs. of 21 variables:
## $ State : Factor w/ 51 levels "AK","AL","AR",..: 17 36 32 36 37 2 20 25 19 50 ...
## $ Account_Length : int 128 107 137 84 75 118 121 147 117 141 ...
## $ Area_Code : int 415 415 415 408 415 510 510 415 408 415 ...
## $ Phone_No : int 3824657 3717191 3581921 3759999 3306626 3918027 3559993 3299001 3354719 3308173 ...
## $ International_Plan : Factor w/ 3 levels ""," no"," yes": 2 2 2 3 3 3 2 3 2 3 ...
## $ Voice_Mail_Plan : Factor w/ 3 levels ""," no"," yes": 3 3 2 2 2 2 3 2 2 3 ...
## $ No_Vmail_Messages : int 25 26 0 0 0 0 24 0 0 37 ...
## $ Total_Day_minutes : num 265 162 243 299 167 ...
## $ Total_Day_Calls : int 110 123 114 71 113 98 88 79 97 84 ...
## $ Total_Day_charge : num 45.1 27.5 41.4 50.9 28.3 ...
## $ Total_Eve_Minutes : num 197.4 195.5 121.2 61.9 148.3 ...
## $ Total_Eve_Calls : int 99 103 110 88 122 101 108 94 80 111 ...
## $ Total_Eve_Charge : num 16.78 16.62 10.3 5.26 12.61 ...
## $ Total_Night_Minutes: num 245 254 163 197 187 ...
## $ Total_Night_Calls : int 91 103 104 89 121 118 118 96 90 97 ...
## $ Total_Night_Charge : num 11.01 11.45 7.32 8.86 8.41 ...
## $ Total_Intl_Minutes : num 10 13.7 12.2 6.6 10.1 6.3 7.5 7.1 8.7 11.2 ...
## $ Total_Intl_Calls : int 3 3 5 7 3 6 7 6 4 5 ...
## $ Total_Intl_Charge : num 2.7 3.7 3.29 1.78 2.73 1.7 2.03 1.92 2.35 3.02 ...
## $ No_CS_Calls : int 1 1 0 2 3 0 3 0 1 0 ...
## $ Churn : logi FALSE FALSE FALSE FALSE FALSE FALSE ...# view the summary of the df
summary(df)## State Account_Length Area_Code Phone_No
## WV : 122 Min. : 1 Min. :408.0 Min. :3271058
## MN : 94 1st Qu.: 74 1st Qu.:415.0 1st Qu.:3511101
## AL : 93 Median :100 Median :415.0 Median :3752748
## VA : 91 Mean :101 Mean :437.3 Mean :3748047
## NY : 89 3rd Qu.:127 3rd Qu.:510.0 3rd Qu.:3987837
## OH : 88 Max. :243 Max. :510.0 Max. :4229964
## (Other):3181 NA's :1 NA's :1
## International_Plan Voice_Mail_Plan No_Vmail_Messages Total_Day_minutes
## : 5 : 4 Min. : 0.000 Min. : 0.0
## no :3390 no :2726 1st Qu.: 0.000 1st Qu.:143.7
## yes: 363 yes:1028 Median : 0.000 Median :179.4
## Mean : 8.026 Mean :180.0
## 3rd Qu.:19.000 3rd Qu.:216.7
## Max. :51.000 Max. :350.8
## NA's :7
## Total_Day_Calls Total_Day_charge Total_Eve_Minutes Total_Eve_Calls
## Min. : 0.0 Min. : 0.00 Min. : 0.0 Min. : 0.0
## 1st Qu.: 87.0 1st Qu.:24.42 1st Qu.:166.4 1st Qu.: 87.0
## Median :101.0 Median :30.50 Median :201.0 Median :100.0
## Mean :100.4 Mean :30.60 Mean :200.8 Mean :100.2
## 3rd Qu.:114.0 3rd Qu.:36.80 3rd Qu.:235.1 3rd Qu.:114.0
## Max. :165.0 Max. :59.64 Max. :363.7 Max. :170.0
## NA's :4 NA's :11 NA's :11 NA's :4
## Total_Eve_Charge Total_Night_Minutes Total_Night_Calls Total_Night_Charge
## Min. : 0.00 Min. : 23.2 Min. : 33.0 Min. : 1.040
## 1st Qu.:14.17 1st Qu.:167.1 1st Qu.: 87.0 1st Qu.: 7.520
## Median :17.10 Median :201.1 Median :100.0 Median : 9.050
## Mean :17.08 Mean :200.8 Mean :100.1 Mean : 9.041
## 3rd Qu.:19.98 3rd Qu.:235.3 3rd Qu.:113.0 3rd Qu.:10.590
## Max. :30.91 Max. :395.0 Max. :175.0 Max. :17.770
## NA's :5 NA's :11 NA's :6 NA's :4
## Total_Intl_Minutes Total_Intl_Calls Total_Intl_Charge No_CS_Calls
## Min. : 0.00 Min. : 0.000 Min. :0.000 Min. :0.000
## 1st Qu.: 8.50 1st Qu.: 3.000 1st Qu.:2.300 1st Qu.:1.000
## Median :10.30 Median : 4.000 Median :2.780 Median :1.000
## Mean :10.25 Mean : 4.471 Mean :2.769 Mean :1.563
## 3rd Qu.:12.10 3rd Qu.: 6.000 3rd Qu.:3.270 3rd Qu.:2.000
## Max. :20.00 Max. :20.000 Max. :5.400 Max. :9.000
## NA's :3 NA's :5
## Churn
## Mode :logical
## FALSE:2850
## TRUE :483
## NA's :425
##
##
##
Data PreProcessing
Create a vector for categorical and numerical data respectively
cat_data <- names(df[, c("State", "Area_Code", "Phone_No", "International_Plan",
"Voice_Mail_Plan", "Churn")])
num_data <- setdiff(names(df), c(cat_data)) View Categorical Data vector for validation purpose
cat_data ## [1] "State" "Area_Code" "Phone_No"
## [4] "International_Plan" "Voice_Mail_Plan" "Churn"View Numerical Data vector for validation purpose
num_data ## [1] "Account_Length" "No_Vmail_Messages" "Total_Day_minutes"
## [4] "Total_Day_Calls" "Total_Day_charge" "Total_Eve_Minutes"
## [7] "Total_Eve_Calls" "Total_Eve_Charge" "Total_Night_Minutes"
## [10] "Total_Night_Calls" "Total_Night_Charge" "Total_Intl_Minutes"
## [13] "Total_Intl_Calls" "Total_Intl_Charge" "No_CS_Calls"Changing Data Types
Changing all numerical and integer data types to numerical, categorical to factor
df <- df %>%
mutate_at(num_data, as.double) %>%
mutate_at(cat_data, as.factor)View Structure and Summary of the dataframe, to make sure the data type conversion is successful
# view structure of the data
str(df)## 'data.frame': 3758 obs. of 21 variables:
## $ State : Factor w/ 51 levels "AK","AL","AR",..: 17 36 32 36 37 2 20 25 19 50 ...
## $ Account_Length : num 128 107 137 84 75 118 121 147 117 141 ...
## $ Area_Code : Factor w/ 3 levels "408","415","510": 2 2 2 1 2 3 3 2 1 2 ...
## $ Phone_No : Factor w/ 3758 levels "3271058","3271319",..: 2157 1762 1249 1910 125 2543 1172 93 324 133 ...
## $ International_Plan : Factor w/ 3 levels ""," no"," yes": 2 2 2 3 3 3 2 3 2 3 ...
## $ Voice_Mail_Plan : Factor w/ 3 levels ""," no"," yes": 3 3 2 2 2 2 3 2 2 3 ...
## $ No_Vmail_Messages : num 25 26 0 0 0 0 24 0 0 37 ...
## $ Total_Day_minutes : num 265 162 243 299 167 ...
## $ Total_Day_Calls : num 110 123 114 71 113 98 88 79 97 84 ...
## $ Total_Day_charge : num 45.1 27.5 41.4 50.9 28.3 ...
## $ Total_Eve_Minutes : num 197.4 195.5 121.2 61.9 148.3 ...
## $ Total_Eve_Calls : num 99 103 110 88 122 101 108 94 80 111 ...
## $ Total_Eve_Charge : num 16.78 16.62 10.3 5.26 12.61 ...
## $ Total_Night_Minutes: num 245 254 163 197 187 ...
## $ Total_Night_Calls : num 91 103 104 89 121 118 118 96 90 97 ...
## $ Total_Night_Charge : num 11.01 11.45 7.32 8.86 8.41 ...
## $ Total_Intl_Minutes : num 10 13.7 12.2 6.6 10.1 6.3 7.5 7.1 8.7 11.2 ...
## $ Total_Intl_Calls : num 3 3 5 7 3 6 7 6 4 5 ...
## $ Total_Intl_Charge : num 2.7 3.7 3.29 1.78 2.73 1.7 2.03 1.92 2.35 3.02 ...
## $ No_CS_Calls : num 1 1 0 2 3 0 3 0 1 0 ...
## $ Churn : Factor w/ 2 levels "FALSE","TRUE": 1 1 1 1 1 1 1 1 1 1 ...# view summary of the data
summary(df)## State Account_Length Area_Code Phone_No International_Plan
## WV : 122 Min. : 1 408 : 934 3271058: 1 : 5
## MN : 94 1st Qu.: 74 415 :1871 3271319: 1 no :3390
## AL : 93 Median :100 510 : 952 3272040: 1 yes: 363
## VA : 91 Mean :101 NA's: 1 3272475: 1
## NY : 89 3rd Qu.:127 3273053: 1
## OH : 88 Max. :243 3273587: 1
## (Other):3181 NA's :1 (Other):3752
## Voice_Mail_Plan No_Vmail_Messages Total_Day_minutes Total_Day_Calls
## : 4 Min. : 0.000 Min. : 0.0 Min. : 0.0
## no :2726 1st Qu.: 0.000 1st Qu.:143.7 1st Qu.: 87.0
## yes:1028 Median : 0.000 Median :179.4 Median :101.0
## Mean : 8.026 Mean :180.0 Mean :100.4
## 3rd Qu.:19.000 3rd Qu.:216.7 3rd Qu.:114.0
## Max. :51.000 Max. :350.8 Max. :165.0
## NA's :7 NA's :4
## Total_Day_charge Total_Eve_Minutes Total_Eve_Calls Total_Eve_Charge
## Min. : 0.00 Min. : 0.0 Min. : 0.0 Min. : 0.00
## 1st Qu.:24.42 1st Qu.:166.4 1st Qu.: 87.0 1st Qu.:14.17
## Median :30.50 Median :201.0 Median :100.0 Median :17.10
## Mean :30.60 Mean :200.8 Mean :100.2 Mean :17.08
## 3rd Qu.:36.80 3rd Qu.:235.1 3rd Qu.:114.0 3rd Qu.:19.98
## Max. :59.64 Max. :363.7 Max. :170.0 Max. :30.91
## NA's :11 NA's :11 NA's :4 NA's :5
## Total_Night_Minutes Total_Night_Calls Total_Night_Charge Total_Intl_Minutes
## Min. : 23.2 Min. : 33.0 Min. : 1.040 Min. : 0.00
## 1st Qu.:167.1 1st Qu.: 87.0 1st Qu.: 7.520 1st Qu.: 8.50
## Median :201.1 Median :100.0 Median : 9.050 Median :10.30
## Mean :200.8 Mean :100.1 Mean : 9.041 Mean :10.25
## 3rd Qu.:235.3 3rd Qu.:113.0 3rd Qu.:10.590 3rd Qu.:12.10
## Max. :395.0 Max. :175.0 Max. :17.770 Max. :20.00
## NA's :11 NA's :6 NA's :4 NA's :3
## Total_Intl_Calls Total_Intl_Charge No_CS_Calls Churn
## Min. : 0.000 Min. :0.000 Min. :0.000 FALSE:2850
## 1st Qu.: 3.000 1st Qu.:2.300 1st Qu.:1.000 TRUE : 483
## Median : 4.000 Median :2.780 Median :1.000 NA's : 425
## Mean : 4.471 Mean :2.769 Mean :1.563
## 3rd Qu.: 6.000 3rd Qu.:3.270 3rd Qu.:2.000
## Max. :20.000 Max. :5.400 Max. :9.000
## NA's :5Missing Values Observation
To observe each variables, find out if there’s any missing values
colSums(is.na(df))## State Account_Length Area_Code Phone_No
## 0 1 1 0
## International_Plan Voice_Mail_Plan No_Vmail_Messages Total_Day_minutes
## 0 0 0 7
## Total_Day_Calls Total_Day_charge Total_Eve_Minutes Total_Eve_Calls
## 4 11 11 4
## Total_Eve_Charge Total_Night_Minutes Total_Night_Calls Total_Night_Charge
## 5 11 6 4
## Total_Intl_Minutes Total_Intl_Calls Total_Intl_Charge No_CS_Calls
## 3 5 0 0
## Churn
## 425Findings and Action on Missing Values Count for each variables:
- Account_Length : 1 -> Impute with Mean
- Area_Code : 1 -> Drop
- Total_Day_minutes : 7 -> Impute with Mean
- Total_Day_Calls : 4 -> Impute with Mean
- Total_Day_charge : 11 -> Impute with Mean
- Total_Eve_Minutes: 11 -> Impute with Mean
- Total_Eve_calls : 4 -> Impute with Mean
- Total_Eve_Charge : 5 -> Impute with Mean
- Total_Night_Minutes : 11 -> Impute with Mean
- Total_Night_Calls : 6 -> Impute with Mean
- Total_Night_Charge : 4 -> Impute with Mean
- Total_Intl_Minutes: 3 -> Impute with Mean
- Total_Intl_Calls : 5 -> Impute with Mean
- Churn : 425 -> Ignore, as it will be splitted later during machine learning models training stage
Drop and Imputation of Missing values
Remove row with missing Area_Code
rm_acc_area_df <- df %>%
filter(is.na(Area_Code) == FALSE)
# Double check on if the data has been removed
colSums(is.na(rm_acc_area_df))## State Account_Length Area_Code Phone_No
## 0 1 0 0
## International_Plan Voice_Mail_Plan No_Vmail_Messages Total_Day_minutes
## 0 0 0 7
## Total_Day_Calls Total_Day_charge Total_Eve_Minutes Total_Eve_Calls
## 4 11 11 4
## Total_Eve_Charge Total_Night_Minutes Total_Night_Calls Total_Night_Charge
## 5 11 6 4
## Total_Intl_Minutes Total_Intl_Calls Total_Intl_Charge No_CS_Calls
## 3 5 0 0
## Churn
## 424Impute the rest of numerical variables with its respective mean value
# 1. Create a list of numerical variables with missing values
list_na <- rm_acc_area_df %>%
select(-c("Churn"))
list_na2 <- colnames(list_na)[ apply(list_na, 2, anyNA) ]
# Create mean
average_missing <- apply(rm_acc_area_df[,colnames(rm_acc_area_df) %in% list_na2],
2,
mean,
na.rm = TRUE)
average_missing## Account_Length Total_Day_minutes Total_Day_Calls Total_Day_charge
## 101.005857 179.997573 100.389289 30.595846
## Total_Eve_Minutes Total_Eve_Calls Total_Eve_Charge Total_Night_Minutes
## 200.790897 100.178257 17.075810 200.851415
## Total_Night_Calls Total_Night_Charge Total_Intl_Minutes Total_Intl_Calls
## 100.060784 9.041071 10.253090 4.471215# Create a new variable with the mean and median
df_replace_missing_values <- rm_acc_area_df %>%
mutate(Total_Day_minutes = ifelse(is.na(Total_Day_minutes), average_missing['Total_Day_minutes'], Total_Day_minutes),
Total_Day_Calls = ifelse(is.na(Total_Day_Calls), average_missing['Total_Day_Calls'], Total_Day_Calls),
Total_Day_charge = ifelse(is.na(Total_Day_charge), average_missing['Total_Day_charge'], Total_Day_charge),
Total_Eve_Minutes = ifelse(is.na(Total_Eve_Minutes), average_missing['Total_Eve_Minutes'], Total_Eve_Minutes),
Total_Eve_Calls = ifelse(is.na(Total_Eve_Calls), average_missing['Total_Eve_Calls'], Total_Eve_Calls),
Total_Eve_Charge = ifelse(is.na(Total_Eve_Charge), average_missing['Total_Eve_Charge'], Total_Eve_Charge),
Total_Night_Minutes = ifelse(is.na(Total_Night_Minutes), average_missing['Total_Night_Minutes'], Total_Night_Minutes),
Total_Night_Calls = ifelse(is.na(Total_Night_Calls), average_missing['Total_Night_Calls'], Total_Night_Calls),
Total_Night_Charge = ifelse(is.na(Total_Night_Charge), average_missing['Total_Night_Charge'], Total_Night_Charge),
Total_Intl_Minutes = ifelse(is.na(Total_Intl_Minutes), average_missing['Total_Intl_Minutes'], Total_Intl_Minutes),
Total_Intl_Calls = ifelse(is.na(Total_Intl_Calls), average_missing['Total_Intl_Calls'], Total_Intl_Calls),
Account_Length = ifelse(is.na(Account_Length), average_missing['Account_Length'], Account_Length)
)
# double check if still have missing values
colSums(is.na(df_replace_missing_values))## State Account_Length Area_Code Phone_No
## 0 0 0 0
## International_Plan Voice_Mail_Plan No_Vmail_Messages Total_Day_minutes
## 0 0 0 0
## Total_Day_Calls Total_Day_charge Total_Eve_Minutes Total_Eve_Calls
## 0 0 0 0
## Total_Eve_Charge Total_Night_Minutes Total_Night_Calls Total_Night_Charge
## 0 0 0 0
## Total_Intl_Minutes Total_Intl_Calls Total_Intl_Charge No_CS_Calls
## 0 0 0 0
## Churn
## 424Detect and Removal of Outliers
Observe Outliers using Boxplot
boxplot(df_replace_missing_values[,c(num_data)])
Remove outliers based on percentile: on 1st and 99th Percentile, the top 1% percentile was removed here
rm_outliers <- df_replace_missing_values %>%
filter(No_Vmail_Messages < quantile(df_replace_missing_values$No_Vmail_Messages, 0.99, na.rm = TRUE),
Total_Day_minutes < quantile(df_replace_missing_values$Total_Day_minutes, 0.99, na.rm = TRUE),
Total_Day_Calls < quantile(df_replace_missing_values$Total_Day_Calls, 0.99, na.rm = TRUE),
Total_Day_charge < quantile(df_replace_missing_values$Total_Day_charge, 0.99, na.rm = TRUE),
Total_Eve_Minutes < quantile(df_replace_missing_values$Total_Eve_Minutes, 0.99, na.rm = TRUE),
Total_Eve_Calls < quantile(df_replace_missing_values$Total_Eve_Calls, 0.99, na.rm = TRUE),
Total_Eve_Charge < quantile(df_replace_missing_values$Total_Eve_Charge, 0.99, na.rm = TRUE),
Total_Night_Minutes < quantile(df_replace_missing_values$Total_Night_Minutes ,0.99, na.rm = TRUE),
Total_Night_Calls < quantile(df_replace_missing_values$Total_Night_Calls ,0.99, na.rm = TRUE),
Total_Night_Charge < quantile(df_replace_missing_values$Total_Night_Charge ,0.99, na.rm = TRUE),
Total_Intl_Minutes < quantile(df_replace_missing_values$Total_Intl_Minutes ,0.99, na.rm = TRUE),
Total_Intl_Calls < quantile(df_replace_missing_values$Total_Intl_Calls ,0.99, na.rm = TRUE),
Total_Intl_Charge < quantile(df_replace_missing_values$Total_Intl_Charge ,0.99, na.rm = TRUE),
No_CS_Calls < quantile(df_replace_missing_values$No_CS_Calls ,0.99, na.rm = TRUE))Check the dimensions of before outliers removal vs after outliers removal
#Check the dimensions of before outliers removal vs after outliers removal
dim(rm_outliers)## [1] 3303 21dim(df_replace_missing_values)## [1] 3757 21Findings:
454 rows of outliers row data were removed
Visualization on Number of Customer Service Calls vs Churn
# take only the Churn Data with no missing values
temp_df <- rm_outliers %>% filter(!is.na(Churn))
unique(temp_df$Churn)## [1] FALSE TRUE
## Levels: FALSE TRUE# Find the mean of Churn and No Churn
temp_df2 <- temp_df %>%
group_by(Churn) %>%
summarise(Mean_No_Of_CS_Calls = mean(No_CS_Calls))## `summarise()` ungrouping output (override with `.groups` argument)# Visualize using Bar chart
ggplot(temp_df2, aes(x = Churn, y = Mean_No_Of_CS_Calls)) +
geom_bar(stat = "identity")
Findings:
Customers with more Service Calls made are more likely to Churn. Hence, we are also interested to predict the No. Of Customer Service Calls to make a hypothesis on how likely a customer will churn.
Normalization : To transform data into the same scale. Method used here is Z-Score Normalization
# Z-Score Normalization
pre_proc <- preProcess(rm_outliers %>% select(c(num_data)), method = c('center','scale', 'YeoJohnson'))## Note: Using an external vector in selections is ambiguous.
## ℹ Use `all_of(num_data)` instead of `num_data` to silence this message.
## ℹ See <https://tidyselect.r-lib.org/reference/faq-external-vector.html>.
## This message is displayed once per session.# Transform the data to Z-Score form
transformed_num_df <- predict(pre_proc, rm_outliers)
# view data
head(transformed_num_df)## State Account_Length Area_Code Phone_No International_Plan Voice_Mail_Plan
## 1 KS 0.6859880 415 3824657 no yes
## 2 OH 0.1705578 415 3717191 no yes
## 3 NJ 0.9040278 415 3581921 no no
## 4 OH -0.4067150 408 3759999 yes no
## 5 OK -0.6370261 415 3306626 yes no
## 6 AL 0.4417770 510 3918027 yes no
## No_Vmail_Messages Total_Day_minutes Total_Day_Calls Total_Day_charge
## 1 1.6665326 1.6650638 0.4988108 1.6679404
## 2 1.6699846 -0.3440224 1.2141869 -0.3454267
## 3 -0.5978186 1.2366500 0.7159743 1.2378804
## 4 -0.5978186 2.3487505 -1.4650229 2.3546493
## 5 -0.5978186 -0.2472975 0.6614328 -0.2484118
## 6 -0.5978186 0.8448907 -0.1361516 0.8450450
## Total_Eve_Minutes Total_Eve_Calls Total_Eve_Charge Total_Night_Minutes
## 1 -0.06790512 -0.06234612 -0.07061632 0.92773910
## 2 -0.10678094 0.14914717 -0.10910569 1.13602654
## 3 -1.57236556 0.52496149 -1.57220332 -0.76873477
## 4 -2.64058065 -0.63108841 -2.63578362 -0.07551299
## 5 -1.05121654 1.18528072 -1.05138699 -0.28010693
## 6 0.41160424 0.04309829 0.40836828 0.06885042
## Total_Night_Calls Total_Night_Charge Total_Intl_Minutes Total_Intl_Calls
## 1 -0.4585665 0.92578064 -0.1233752 -0.5326307
## 2 0.1743618 1.13603960 1.3746706 -0.5326307
## 3 0.2276297 -0.76732531 0.7452051 0.4236901
## 4 -0.5628760 -0.07696591 -1.3217243 1.1692059
## 5 1.1445676 -0.28126987 -0.0853833 -0.5326307
## 6 0.9812507 0.06954524 -1.4180320 0.8158933
## Total_Intl_Charge No_CS_Calls Churn
## 1 -0.12894237 -0.2557483 FALSE
## 2 1.37894093 -0.2557483 FALSE
## 3 0.73490717 -1.5050552 FALSE
## 4 -1.31737248 0.6124466 FALSE
## 5 -0.08686774 1.2995078 FALSE
## 6 -1.41097785 -1.5050552 FALSEOne-Hot-Encoding for Categorical Data
To train a regression mode, the data must be numerical, to solve this issue, one hot encoding is used to transformed categorical data into numerical data by flagging the varialble with 1 and 0 as presence and absence.
# one hot encode the categorical variable
cat_data## [1] "State" "Area_Code" "Phone_No"
## [4] "International_Plan" "Voice_Mail_Plan" "Churn"# select features to be one hot encoded (categorical variables)
# Redo this step to ensure all the to be one-hot-encoded data is factor variables
to_onehot <- transformed_num_df %>%
select(c(cat_data)) %>%
mutate_all(as.factor)## Note: Using an external vector in selections is ambiguous.
## ℹ Use `all_of(cat_data)` instead of `cat_data` to silence this message.
## ℹ See <https://tidyselect.r-lib.org/reference/faq-external-vector.html>.
## This message is displayed once per session.head(to_onehot)## State Area_Code Phone_No International_Plan Voice_Mail_Plan Churn
## 1 KS 415 3824657 no yes FALSE
## 2 OH 415 3717191 no yes FALSE
## 3 NJ 415 3581921 no no FALSE
## 4 OH 408 3759999 yes no FALSE
## 5 OK 415 3306626 yes no FALSE
## 6 AL 510 3918027 yes no FALSEstr(to_onehot)## 'data.frame': 3303 obs. of 6 variables:
## $ State : Factor w/ 51 levels "AK","AL","AR",..: 17 36 32 36 37 2 25 16 13 27 ...
## $ Area_Code : Factor w/ 3 levels "408","415","510": 2 2 2 1 2 3 2 2 1 3 ...
## $ Phone_No : Factor w/ 3758 levels "3271058","3271319",..: 2157 1762 1249 1910 125 2543 93 83 1435 2665 ...
## $ International_Plan: Factor w/ 3 levels ""," no"," yes": 2 2 2 3 3 3 3 2 2 2 ...
## $ Voice_Mail_Plan : Factor w/ 3 levels ""," no"," yes": 3 3 2 2 2 2 2 2 2 2 ...
## $ Churn : Factor w/ 2 levels "FALSE","TRUE": 1 1 1 1 1 1 1 2 1 1 ...dmy <- dummyVars(Phone_No ~ ., data = to_onehot)
onehot_df <- data.frame(predict(dmy, newdata = to_onehot)) %>%
cbind(to_onehot %>% select(Phone_No))## Warning in model.frame.default(Terms, newdata, na.action = na.action, xlev =
## object$lvls): variable 'Phone_No' is not a factor# take a look at the data
glimpse(onehot_df)## Rows: 3,303
## Columns: 63
## $ State.AK <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.AL <dbl> 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.AR <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.AZ <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.CA <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.CO <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.CT <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.DC <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.DE <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.FL <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.GA <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.HI <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.IA <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0, …
## $ State.ID <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, …
## $ State.IL <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.IN <dbl> 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, …
## $ State.KS <dbl> 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.KY <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.LA <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.MA <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.MD <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.ME <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.MI <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.MN <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.MO <dbl> 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.MS <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.MT <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, …
## $ State.NC <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.ND <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.NE <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.NH <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.NJ <dbl> 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.NM <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.NV <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.NY <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.OH <dbl> 0, 1, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.OK <dbl> 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.OR <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.PA <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.RI <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.SC <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.SD <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.TN <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.TX <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, …
## $ State.UT <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.VA <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, …
## $ State.VT <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, …
## $ State.WA <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.WI <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.WV <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.WY <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ Area_Code.408 <dbl> 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, …
## $ Area_Code.415 <dbl> 1, 1, 1, 0, 1, 0, 1, 1, 0, 0, 1, 0, 0, 0, 1, …
## $ Area_Code.510 <dbl> 0, 0, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 1, 1, 0, …
## $ International_Plan. <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ International_Plan..no <dbl> 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, …
## $ International_Plan..yes <dbl> 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ Voice_Mail_Plan. <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ Voice_Mail_Plan..no <dbl> 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 0, 1, …
## $ Voice_Mail_Plan..yes <dbl> 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, …
## $ Churn.FALSE <dbl> 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, …
## $ Churn.TRUE <dbl> 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, …
## $ Phone_No <fct> 3824657, 3717191, 3581921, 3759999, 3306626, …Splitting of Classification and Regression dataset & Joining of the one-hot-encoded dataset with the numerical variables
Classification dataset uses the normalized data, while regression dataset uses the unnormalized data for easier checking purpose.
# join numerical and onehot encoded features
# Data set for classification
final_df_class <- transformed_num_df
glimpse(final_df_class)## Rows: 3,303
## Columns: 21
## $ State <fct> KS, OH, NJ, OH, OK, AL, MO, IN, IA, MT, IA, ID, V…
## $ Account_Length <dbl> 0.6859880, 0.1705578, 0.9040278, -0.4067150, -0.6…
## $ Area_Code <fct> 415, 415, 415, 408, 415, 510, 415, 415, 408, 510,…
## $ Phone_No <fct> 3824657, 3717191, 3581921, 3759999, 3306626, 3918…
## $ International_Plan <fct> no, no, no, yes, yes, yes, yes, no, no, …
## $ Voice_Mail_Plan <fct> yes, yes, no, no, no, no, no, no, no, n…
## $ No_Vmail_Messages <dbl> 1.6665326, 1.6699846, -0.5978186, -0.5978186, -0.…
## $ Total_Day_minutes <dbl> 1.6650638, -0.3440224, 1.2366500, 2.3487505, -0.2…
## $ Total_Day_Calls <dbl> 0.4988108, 1.2141869, 0.7159743, -1.4650229, 0.66…
## $ Total_Day_charge <dbl> 1.6679404, -0.3454267, 1.2378804, 2.3546493, -0.2…
## $ Total_Eve_Minutes <dbl> -0.06790512, -0.10678094, -1.57236556, -2.6405806…
## $ Total_Eve_Calls <dbl> -0.062346121, 0.149147168, 0.524961493, -0.631088…
## $ Total_Eve_Charge <dbl> -0.07061632, -0.10910569, -1.57220332, -2.6357836…
## $ Total_Night_Minutes <dbl> 0.92773910, 1.13602654, -0.76873477, -0.07551299,…
## $ Total_Night_Calls <dbl> -0.45856646, 0.17436180, 0.22762974, -0.56287596,…
## $ Total_Night_Charge <dbl> 0.92578064, 1.13603960, -0.76732531, -0.07696591,…
## $ Total_Intl_Minutes <dbl> -0.123375195, 1.374670631, 0.745205132, -1.321724…
## $ Total_Intl_Calls <dbl> -0.53263066, -0.53263066, 0.42369010, 1.16920585,…
## $ Total_Intl_Charge <dbl> -0.12894237, 1.37894093, 0.73490717, -1.31737248,…
## $ No_CS_Calls <dbl> -0.2557483, -0.2557483, -1.5050552, 0.6124466, 1.…
## $ Churn <fct> FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, …# Data set for Regression
final_df_regres <- rm_outliers %>%
left_join(onehot_df, by ='Phone_No')
glimpse(final_df_regres)## Rows: 3,303
## Columns: 83
## $ State <fct> KS, OH, NJ, OH, OK, AL, MO, IN, IA, MT, IA, I…
## $ Account_Length <dbl> 128, 107, 137, 84, 75, 118, 147, 65, 168, 95,…
## $ Area_Code <fct> 415, 415, 415, 408, 415, 510, 415, 415, 408, …
## $ Phone_No <fct> 3824657, 3717191, 3581921, 3759999, 3306626, …
## $ International_Plan <fct> no, no, no, yes, yes, yes, yes, no, …
## $ Voice_Mail_Plan <fct> yes, yes, no, no, no, no, no, no, no…
## $ No_Vmail_Messages <dbl> 25, 26, 0, 0, 0, 0, 0, 0, 0, 0, 0, 27, 0, 33,…
## $ Total_Day_minutes <dbl> 265.1, 161.6, 243.4, 299.4, 166.7, 223.4, 157…
## $ Total_Day_Calls <dbl> 110, 123, 114, 71, 113, 98, 79, 137, 96, 88, …
## $ Total_Day_charge <dbl> 45.07, 27.47, 41.38, 50.90, 28.34, 37.98, 26.…
## $ Total_Eve_Minutes <dbl> 197.4, 195.5, 121.2, 61.9, 148.3, 220.6, 103.…
## $ Total_Eve_Calls <dbl> 99, 103, 110, 88, 122, 101, 94, 83, 71, 75, 7…
## $ Total_Eve_Charge <dbl> 16.78, 16.62, 10.30, 5.26, 12.61, 18.75, 8.76…
## $ Total_Night_Minutes <dbl> 244.7, 254.4, 162.6, 196.9, 186.9, 203.9, 211…
## $ Total_Night_Calls <dbl> 91, 103, 104, 89, 121, 118, 96, 111, 128, 115…
## $ Total_Night_Charge <dbl> 11.01, 11.45, 7.32, 8.86, 8.41, 9.18, 9.53, 9…
## $ Total_Intl_Minutes <dbl> 10.0, 13.7, 12.2, 6.6, 10.1, 6.3, 7.1, 12.7, …
## $ Total_Intl_Calls <dbl> 3, 3, 5, 7, 3, 6, 6, 6, 2, 5, 6, 4, 3, 5, 2, …
## $ Total_Intl_Charge <dbl> 2.70, 3.70, 3.29, 1.78, 2.73, 1.70, 1.92, 3.4…
## $ No_CS_Calls <dbl> 1, 1, 0, 2, 3, 0, 0, 4, 1, 3, 4, 1, 3, 1, 1, …
## $ Churn <fct> FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FAL…
## $ State.AK <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.AL <dbl> 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.AR <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.AZ <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.CA <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.CO <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.CT <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.DC <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.DE <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.FL <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.GA <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.HI <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.IA <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0, …
## $ State.ID <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, …
## $ State.IL <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.IN <dbl> 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, …
## $ State.KS <dbl> 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.KY <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.LA <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.MA <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.MD <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.ME <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.MI <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.MN <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.MO <dbl> 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.MS <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.MT <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, …
## $ State.NC <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.ND <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.NE <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.NH <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.NJ <dbl> 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.NM <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.NV <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.NY <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.OH <dbl> 0, 1, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.OK <dbl> 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.OR <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.PA <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.RI <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.SC <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.SD <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.TN <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.TX <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, …
## $ State.UT <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.VA <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, …
## $ State.VT <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, …
## $ State.WA <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.WI <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.WV <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ State.WY <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ Area_Code.408 <dbl> 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, …
## $ Area_Code.415 <dbl> 1, 1, 1, 0, 1, 0, 1, 1, 0, 0, 1, 0, 0, 0, 1, …
## $ Area_Code.510 <dbl> 0, 0, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 1, 1, 0, …
## $ International_Plan. <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ International_Plan..no <dbl> 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, …
## $ International_Plan..yes <dbl> 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ Voice_Mail_Plan. <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ Voice_Mail_Plan..no <dbl> 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 0, 1, …
## $ Voice_Mail_Plan..yes <dbl> 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, …
## $ Churn.FALSE <dbl> 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, …
## $ Churn.TRUE <dbl> 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, …View the dimensions of both Classification and Regression dataset
dim(final_df_class)## [1] 3303 21dim(final_df_regres)## [1] 3303 83
Modeling
Duplicate both of the cleaned dataframes to avoid corruption during modeling
final_df_class1 <- data.frame(final_df_class)
final_df_regres1 <- data.frame(final_df_regres)Final data cleaning for classification modeling:
For this classification problem, the target variable to be classified by the model is the Churn variable which is basically the industry term for customer stopping the use or subscription of a company’s services, in this case it is a telco service provider. Hence, the objective is to try to classify customers that will churn based on thier day-to-day usage such as total minutes of calls everday, subscription of international plan, total day charge and so on. There are still some NA values which cannot be imputed or filled as it is the target variable, which are not removed earlier on as the observations with NA target values can be used in the proceeding regression problem which has another variable as target value. The observations with na values for the target variable are removed.
#Find any na values
colSums(is.na(final_df_class1))## State Account_Length Area_Code Phone_No
## 0 0 0 0
## International_Plan Voice_Mail_Plan No_Vmail_Messages Total_Day_minutes
## 0 0 0 0
## Total_Day_Calls Total_Day_charge Total_Eve_Minutes Total_Eve_Calls
## 0 0 0 0
## Total_Eve_Charge Total_Night_Minutes Total_Night_Calls Total_Night_Charge
## 0 0 0 0
## Total_Intl_Minutes Total_Intl_Calls Total_Intl_Charge No_CS_Calls
## 0 0 0 0
## Churn
## 371head(final_df_class1)## State Account_Length Area_Code Phone_No International_Plan Voice_Mail_Plan
## 1 KS 0.6859880 415 3824657 no yes
## 2 OH 0.1705578 415 3717191 no yes
## 3 NJ 0.9040278 415 3581921 no no
## 4 OH -0.4067150 408 3759999 yes no
## 5 OK -0.6370261 415 3306626 yes no
## 6 AL 0.4417770 510 3918027 yes no
## No_Vmail_Messages Total_Day_minutes Total_Day_Calls Total_Day_charge
## 1 1.6665326 1.6650638 0.4988108 1.6679404
## 2 1.6699846 -0.3440224 1.2141869 -0.3454267
## 3 -0.5978186 1.2366500 0.7159743 1.2378804
## 4 -0.5978186 2.3487505 -1.4650229 2.3546493
## 5 -0.5978186 -0.2472975 0.6614328 -0.2484118
## 6 -0.5978186 0.8448907 -0.1361516 0.8450450
## Total_Eve_Minutes Total_Eve_Calls Total_Eve_Charge Total_Night_Minutes
## 1 -0.06790512 -0.06234612 -0.07061632 0.92773910
## 2 -0.10678094 0.14914717 -0.10910569 1.13602654
## 3 -1.57236556 0.52496149 -1.57220332 -0.76873477
## 4 -2.64058065 -0.63108841 -2.63578362 -0.07551299
## 5 -1.05121654 1.18528072 -1.05138699 -0.28010693
## 6 0.41160424 0.04309829 0.40836828 0.06885042
## Total_Night_Calls Total_Night_Charge Total_Intl_Minutes Total_Intl_Calls
## 1 -0.4585665 0.92578064 -0.1233752 -0.5326307
## 2 0.1743618 1.13603960 1.3746706 -0.5326307
## 3 0.2276297 -0.76732531 0.7452051 0.4236901
## 4 -0.5628760 -0.07696591 -1.3217243 1.1692059
## 5 1.1445676 -0.28126987 -0.0853833 -0.5326307
## 6 0.9812507 0.06954524 -1.4180320 0.8158933
## Total_Intl_Charge No_CS_Calls Churn
## 1 -0.12894237 -0.2557483 FALSE
## 2 1.37894093 -0.2557483 FALSE
## 3 0.73490717 -1.5050552 FALSE
## 4 -1.31737248 0.6124466 FALSE
## 5 -0.08686774 1.2995078 FALSE
## 6 -1.41097785 -1.5050552 FALSE#Remove na values
final_df_class1 <- final_df_class1[complete.cases(final_df_class1), ]
#Check for na values again
colSums(is.na(final_df_class1))## State Account_Length Area_Code Phone_No
## 0 0 0 0
## International_Plan Voice_Mail_Plan No_Vmail_Messages Total_Day_minutes
## 0 0 0 0
## Total_Day_Calls Total_Day_charge Total_Eve_Minutes Total_Eve_Calls
## 0 0 0 0
## Total_Eve_Charge Total_Night_Minutes Total_Night_Calls Total_Night_Charge
## 0 0 0 0
## Total_Intl_Minutes Total_Intl_Calls Total_Intl_Charge No_CS_Calls
## 0 0 0 0
## Churn
## 0#Check data types for each varable
str(final_df_class1)## 'data.frame': 2932 obs. of 21 variables:
## $ State : Factor w/ 51 levels "AK","AL","AR",..: 17 36 32 36 37 2 25 16 13 27 ...
## $ Account_Length : num 0.686 0.171 0.904 -0.407 -0.637 ...
## $ Area_Code : Factor w/ 3 levels "408","415","510": 2 2 2 1 2 3 2 2 1 3 ...
## $ Phone_No : Factor w/ 3758 levels "3271058","3271319",..: 2157 1762 1249 1910 125 2543 93 83 1435 2665 ...
## $ International_Plan : Factor w/ 3 levels ""," no"," yes": 2 2 2 3 3 3 3 2 2 2 ...
## $ Voice_Mail_Plan : Factor w/ 3 levels ""," no"," yes": 3 3 2 2 2 2 2 2 2 2 ...
## $ No_Vmail_Messages : num 1.667 1.67 -0.598 -0.598 -0.598 ...
## $ Total_Day_minutes : num 1.665 -0.344 1.237 2.349 -0.247 ...
## $ Total_Day_Calls : num 0.499 1.214 0.716 -1.465 0.661 ...
## $ Total_Day_charge : num 1.668 -0.345 1.238 2.355 -0.248 ...
## $ Total_Eve_Minutes : num -0.0679 -0.1068 -1.5724 -2.6406 -1.0512 ...
## $ Total_Eve_Calls : num -0.0623 0.1491 0.525 -0.6311 1.1853 ...
## $ Total_Eve_Charge : num -0.0706 -0.1091 -1.5722 -2.6358 -1.0514 ...
## $ Total_Night_Minutes: num 0.9277 1.136 -0.7687 -0.0755 -0.2801 ...
## $ Total_Night_Calls : num -0.459 0.174 0.228 -0.563 1.145 ...
## $ Total_Night_Charge : num 0.926 1.136 -0.767 -0.077 -0.281 ...
## $ Total_Intl_Minutes : num -0.1234 1.3747 0.7452 -1.3217 -0.0854 ...
## $ Total_Intl_Calls : num -0.533 -0.533 0.424 1.169 -0.533 ...
## $ Total_Intl_Charge : num -0.1289 1.3789 0.7349 -1.3174 -0.0869 ...
## $ No_CS_Calls : num -0.256 -0.256 -1.505 0.612 1.3 ...
## $ Churn : Factor w/ 2 levels "FALSE","TRUE": 1 1 1 1 1 1 1 2 1 1 ...Split the dataset by 80-20, 80% for training and 20% for testing
validation_index <- createDataPartition(final_df_class1$Churn, p=0.80, list=FALSE)
final_df_class_test <- final_df_class1[-validation_index,]
final_df_class_train <- final_df_class1[validation_index,]
dim(final_df_class_train)## [1] 2347 21dim(final_df_class_test)## [1] 585 21Setup and run the Random Forest algorithm on the training set:
3 fold cross validation applied to guage the overall training accuracy. Seed is set to 7 as a pseudo random number generator to ensure the accuracy produced is the same everytime.
# Run algorithms using 3-fold cross validation
method1 <- trainControl(method="cv", number=3)
# Random Forest
set.seed(7)
fit.rf <- train(Churn~., data=final_df_class_train, method="ranger", metric='Accuracy', trControl=method1)## Growing trees.. Progress: 29%. Estimated remaining time: 1 minute, 17 seconds.
## Growing trees.. Progress: 59%. Estimated remaining time: 43 seconds.
## Growing trees.. Progress: 92%. Estimated remaining time: 7 seconds.
## Growing trees.. Progress: 27%. Estimated remaining time: 1 minute, 24 seconds.
## Growing trees.. Progress: 55%. Estimated remaining time: 51 seconds.
## Growing trees.. Progress: 83%. Estimated remaining time: 18 seconds.
## Growing trees.. Progress: 35%. Estimated remaining time: 57 seconds.
## Growing trees.. Progress: 70%. Estimated remaining time: 26 seconds.
## Growing trees.. Progress: 28%. Estimated remaining time: 1 minute, 20 seconds.
## Growing trees.. Progress: 55%. Estimated remaining time: 49 seconds.
## Growing trees.. Progress: 83%. Estimated remaining time: 19 seconds.
## Growing trees.. Progress: 31%. Estimated remaining time: 1 minute, 8 seconds.
## Growing trees.. Progress: 64%. Estimated remaining time: 35 seconds.
## Growing trees.. Progress: 97%. Estimated remaining time: 3 seconds.
## Growing trees.. Progress: 26%. Estimated remaining time: 1 minute, 26 seconds.
## Growing trees.. Progress: 54%. Estimated remaining time: 52 seconds.
## Growing trees.. Progress: 80%. Estimated remaining time: 22 seconds.
## Growing trees.. Progress: 14%. Estimated remaining time: 3 minutes, 16 seconds.
## Growing trees.. Progress: 28%. Estimated remaining time: 2 minutes, 40 seconds.
## Growing trees.. Progress: 42%. Estimated remaining time: 2 minutes, 13 seconds.
## Growing trees.. Progress: 56%. Estimated remaining time: 1 minute, 39 seconds.
## Growing trees.. Progress: 69%. Estimated remaining time: 1 minute, 12 seconds.
## Growing trees.. Progress: 83%. Estimated remaining time: 39 seconds.
## Growing trees.. Progress: 97%. Estimated remaining time: 5 seconds.fit.rf## Random Forest
##
## 2347 samples
## 20 predictor
## 2 classes: 'FALSE', 'TRUE'
##
## No pre-processing
## Resampling: Cross-Validated (3 fold)
## Summary of sample sizes: 1564, 1564, 1566
## Resampling results across tuning parameters:
##
## mtry splitrule Accuracy Kappa
## 2 gini 0.8789950 0.0000000
## 2 extratrees 0.8789950 0.0000000
## 87 gini 0.8956174 0.2170439
## 87 extratrees 0.8789950 0.0000000
## 3827 gini 0.9454684 0.7039203
## 3827 extratrees 0.9467444 0.7019253
##
## Tuning parameter 'min.node.size' was held constant at a value of 1
## Accuracy was used to select the optimal model using the largest value.
## The final values used for the model were mtry = 3827, splitrule =
## extratrees and min.node.size = 1.Plot the learning curve of the algorithm training process:
The plot shows the number of randomly selected predictors at each split and the respective accuracy achieved. The accuracy is the best when the number of of randomly selected predictors at each split is 3827. Splitting rule of extratrees performed slightly better than the gini rule.
plot(fit.rf)
Apply the built Random Forest model on the testing set and show the predicton performance
An accuracy of more than 90% is achieved on the test set, which is very close to what has been achieved on the training set which means that the model is able to generalize well and there is no issue of overfitting or underfitting. With that, the model is able to classify more than 500 non-churn observations out of a total of 539, and it classified close to all non-churn observations correctly.
#Test the RF model on the test set
predictions <- predict(fit.rf, final_df_class_test)
#Show the prediction stats on the test set
confusionMatrix(predictions, final_df_class_test$Churn)## Confusion Matrix and Statistics
##
## Reference
## Prediction FALSE TRUE
## FALSE 508 18
## TRUE 7 52
##
## Accuracy : 0.9573
## 95% CI : (0.9376, 0.9722)
## No Information Rate : 0.8803
## P-Value [Acc > NIR] : 7.659e-11
##
## Kappa : 0.7824
##
## Mcnemar's Test P-Value : 0.0455
##
## Sensitivity : 0.9864
## Specificity : 0.7429
## Pos Pred Value : 0.9658
## Neg Pred Value : 0.8814
## Prevalence : 0.8803
## Detection Rate : 0.8684
## Detection Prevalence : 0.8991
## Balanced Accuracy : 0.8646
##
## 'Positive' Class : FALSE
##
Final data cleaning for regression modeling:
Split the dataset by 80-20, 80% for training and 20% for testing
validation_index <- createDataPartition(final_df_regres1$No_CS_Calls, p=0.80, list=FALSE)
final_df_regres_test <- final_df_regres1[-validation_index,]
final_df_regres_train <- final_df_regres1[validation_index,]
dim(final_df_regres_train)## [1] 2643 80dim(final_df_regres_test)## [1] 660 80Setup and run the neural network algorithm (Multilayer Perceptron, MLP) on the training set:
3 fold cross validation applied to guage the overall training accuracy. Seed is set to 7 as a pseudo random number generator to ensure the accuracy produced is the same everytime.
# Run algorithms using 3-fold cross validation
method1 <- trainControl(method="cv", number=3)
# Regression with Neural Network model(Multilayer Perceptron model)
set.seed(7)
fit.mlp <- train(No_CS_Calls~., data=final_df_regres_train, method="mlp", metric='MAE', trControl=method1)## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)## Warning in nominalTrainWorkflow(x = x, y = y, wts = weights, info = trainInfo, :
## There were missing values in resampled performance measures.## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)
## Warning in snnsObject$setUnitName(num, iNames[[i]]): SNNS error message in
## setUnitName : SNNS-Kernel Error: Symbol pattern invalid (must match [A-Za-z]
## [^|, ]*)fit.mlp## Multi-Layer Perceptron
##
## 2643 samples
## 79 predictor
##
## No pre-processing
## Resampling: Cross-Validated (3 fold)
## Summary of sample sizes: 1762, 1761, 1763
## Resampling results across tuning parameters:
##
## size RMSE Rsquared MAE
## 1 1.661514 0.0001705958 1.343567
## 3 1.470485 0.0014212059 1.146664
## 5 1.374573 NaN 1.137586
##
## MAE was used to select the optimal model using the smallest value.
## The final value used for the model was size = 5.Plot the learning curve of the MLP algorithm training process:
Apply the built MLP model on the testing set and show the predicton performance
A MAE of 1.16 is achieved on the test set, which is only slightly higher of an prediction error than what has been achieved on the training set which means that the model is able to generalize well and there is no major issue of overfitting or underfitting.
#Test the MLP model on the test set
predictions <- predict(fit.mlp, final_df_regres_test)
#Create function to calculate RMSE for test set
calc_rmse = function(actual, predicted) {
sqrt(mean((actual - predicted) ^ 2))
}
#Show the RMSE on the test set
calc_rmse(actual = final_df_regres_test$No_CS_Calls,
predicted = predict(fit.mlp, final_df_regres_test))## [1] 1.162633
Conclusion
As conclusion, both modeling for the subjected classification and regression problems using machine learning models have been done with relatively well prediction accuracies. For classification, a Random Forest model has been used to predict customer churn, an accuracy of 95.9% on the testing dataset has been achieved. Whereas for the regression problem, a Neural Network model has been used in the form of a Multilayer Perceptron model, the number of customer service calls made has been predicted with a Mean Absolute Error of 1.16 (No. of Calls) which is relatively acceptable. With that, the telco operator should be able to effectively identify customers who are likely to churn as well as customers who are more likely to be dissatisfied, to ultimately come up with potential solution to increase customer retention as well as service quality.