Fish Aggressiveness prediction project
Adekunle Joseph Damilare
2024-04-17
Preliminaries
set.seed(10)
library(haven)
library(tidyverse)## Warning: package 'ggplot2' was built under R version 4.3.3## ── Attaching core tidyverse packages ──────────────────────── tidyverse 2.0.0 ──
## ✔ dplyr 1.1.3 ✔ readr 2.1.4
## ✔ forcats 1.0.0 ✔ stringr 1.5.0
## ✔ ggplot2 3.5.0 ✔ tibble 3.2.1
## ✔ lubridate 1.9.3 ✔ tidyr 1.3.0
## ✔ purrr 1.0.2
## ── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──
## ✖ dplyr::filter() masks stats::filter()
## ✖ dplyr::lag() masks stats::lag()
## ℹ Use the conflicted package (<http://conflicted.r-lib.org/>) to force all conflicts to become errorstransformed_boluwatife <- read_sav("C:/Users/DELL/Desktop/2024_Projects/project work from boluwatife/transformed_boluwatife.sav")
head(transformed_boluwatife,5)## # A tibble: 5 × 9
## Behaviour Group Week_0 Week_3 Week_5 Week_7 Week_9 Week_11 Week_13
## <dbl+lbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 1 [SL] 1 6.8 5.6 6.2 5.2 5.2 7.1 5.5
## 2 1 [SL] 1 6.4 7 5.9 4.8 5.8 6.2 5.1
## 3 1 [SL] 1 4.6 6.1 5 5.1 8.1 7.4 6.1
## 4 1 [SL] 1 7.5 5.8 4.9 5.2 6.4 5.8 5.9
## 5 1 [SL] 1 5.2 4.9 5 6.6 6.3 6 7.5Data Cleaning
transformed_boluwatife$Group <- ifelse(transformed_boluwatife$Group== 1, 0, 1)Labeling Behaviour
ucl_decoded <- transformed_boluwatife %>%
mutate(Behaviour = case_when(
Behaviour == 1 ~ "SL",
Behaviour == 2 ~ " TL",
Behaviour == 3 ~ "Aggressive",
Behaviour == 4 ~ "Weight"
))
head(ucl_decoded, 10)## # A tibble: 10 × 9
## Behaviour Group Week_0 Week_3 Week_5 Week_7 Week_9 Week_11 Week_13
## <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 SL 0 6.8 5.6 6.2 5.2 5.2 7.1 5.5
## 2 SL 0 6.4 7 5.9 4.8 5.8 6.2 5.1
## 3 SL 0 4.6 6.1 5 5.1 8.1 7.4 6.1
## 4 SL 0 7.5 5.8 4.9 5.2 6.4 5.8 5.9
## 5 SL 0 5.2 4.9 5 6.6 6.3 6 7.5
## 6 SL 0 5 5.1 4.6 7.2 5.5 5 7.7
## 7 SL 0 5.6 6.2 7.9 5.7 7 6.8 6.6
## 8 SL 0 4.6 4.8 5 4.8 7.3 5.4 5.4
## 9 SL 0 6.5 6.9 6.7 6.9 6.1 5 5.1
## 10 SL 0 4.2 4.9 6.4 8 6.2 5.3 6.9Data Exploration
library(dlookr)## Warning: package 'dlookr' was built under R version 4.4.0## Registered S3 methods overwritten by 'dlookr':
## method from
## plot.transform scales
## print.transform scales##
## Attaching package: 'dlookr'## The following object is masked from 'package:tidyr':
##
## extract## The following object is masked from 'package:base':
##
## transformChecking for Missing values
Checking for outliers
diagnose_outlier(transformed_boluwatife)## # A tibble: 8 × 6
## variables outliers_cnt outliers_ratio outliers_mean with_mean without_mean
## <chr> <int> <dbl> <dbl> <dbl> <dbl>
## 1 Group 0 0 NaN 0.5 0.5
## 2 Week_0 60 18.8 21.5 8.23 5.17
## 3 Week_3 55 17.2 22.6 8.25 5.28
## 4 Week_5 59 18.4 22.9 8.54 5.29
## 5 Week_7 58 18.1 24.1 8.87 5.50
## 6 Week_9 48 15 24.0 8.59 5.88
## 7 Week_11 41 12.8 24 8.51 6.24
## 8 Week_13 52 16.2 26.2 9.22 5.92library(visdat)## Warning: package 'visdat' was built under R version 4.3.3Checking for data structure
Correlation Analysis
library(corrplot)## Warning: package 'corrplot' was built under R version 4.3.2## corrplot 0.92 loadedcr_data <- transformed_boluwatife %>% select(-c(Group, Behaviour))
coq <- cor(cr_data)
corrplot(coq, method = "number")
plot(cr_data, col ="red")
Descriptive analysis
library(summarytools)## Warning: package 'summarytools' was built under R version 4.3.3##
## Attaching package: 'summarytools'## The following object is masked from 'package:tibble':
##
## viewdfSummary(cr_data$Week_0)## cr_data$Week_0 was converted to a data frame## Data Frame Summary
## cr_data
## Dimensions: 320 x 1
## Duplicates: 196
##
## ------------------------------------------------------------------------------------------------------------
## No Variable Label Stats / Values Freqs (% of Valid) Graph Valid Missing
## ---- ----------- ----------- ---------------------- --------------------- ------------- ---------- ---------
## 1 Week_0 Week Zero Mean (sd) : 8.2 (7) 124 distinct values : 320 0
## [numeric] min < med < max: : . (100.0%) (0.0%)
## 2.2 < 5.2 < 34 : :
## IQR (CV) : 3.6 (0.8) : :
## : : . . . .
## ------------------------------------------------------------------------------------------------------------dfSummary(cr_data$Week_3)## cr_data$Week_3 was converted to a data frame## Data Frame Summary
## cr_data
## Dimensions: 320 x 1
## Duplicates: 199
##
## ----------------------------------------------------------------------------------------------------------------
## No Variable Label Stats / Values Freqs (% of Valid) Graph Valid Missing
## ---- ----------- ------------ ----------------------- --------------------- --------------- ---------- ---------
## 1 Week_3 Week Three Mean (sd) : 8.3 (7.3) 121 distinct values : 320 0
## [numeric] min < med < max: : . (100.0%) (0.0%)
## 2 < 5.2 < 37 : :
## IQR (CV) : 4 (0.9) : :
## : : . . . . .
## ----------------------------------------------------------------------------------------------------------------dfSummary(cr_data$Week_5)## cr_data$Week_5 was converted to a data frame## Data Frame Summary
## cr_data
## Dimensions: 320 x 1
## Duplicates: 197
##
## -------------------------------------------------------------------------------------------------------------
## No Variable Label Stats / Values Freqs (% of Valid) Graph Valid Missing
## ---- ----------- ----------- ----------------------- --------------------- ------------- ---------- ---------
## 1 Week_5 Week Five Mean (sd) : 8.5 (7.6) 123 distinct values : 320 0
## [numeric] min < med < max: : . (100.0%) (0.0%)
## 2 < 5.3 < 35 : :
## IQR (CV) : 4.1 (0.9) : :
## : : . : . .
## -------------------------------------------------------------------------------------------------------------dfSummary(cr_data$Week_7)## cr_data$Week_7 was converted to a data frame## Data Frame Summary
## cr_data
## Dimensions: 320 x 1
## Duplicates: 185
##
## --------------------------------------------------------------------------------------------------------------
## No Variable Label Stats / Values Freqs (% of Valid) Graph Valid Missing
## ---- ----------- ------------ ----------------------- --------------------- ------------- ---------- ---------
## 1 Week_7 Week Seven Mean (sd) : 8.9 (7.9) 135 distinct values : 320 0
## [numeric] min < med < max: : : (100.0%) (0.0%)
## 2 < 5.3 < 36 : :
## IQR (CV) : 4.4 (0.9) : :
## : : . . . .
## --------------------------------------------------------------------------------------------------------------dfSummary(cr_data$Week_9)## cr_data$Week_9 was converted to a data frame## Data Frame Summary
## cr_data
## Dimensions: 320 x 1
## Duplicates: 180
##
## -------------------------------------------------------------------------------------------------------------
## No Variable Label Stats / Values Freqs (% of Valid) Graph Valid Missing
## ---- ----------- ----------- ----------------------- --------------------- ------------- ---------- ---------
## 1 Week_9 Week Nine Mean (sd) : 8.6 (7.3) 140 distinct values : 320 0
## [numeric] min < med < max: : : (100.0%) (0.0%)
## 2 < 5.3 < 33 : :
## IQR (CV) : 4.8 (0.9) : :
## : : : . . .
## -------------------------------------------------------------------------------------------------------------dfSummary(cr_data$Week_11)## cr_data$Week_11 was converted to a data frame## Data Frame Summary
## cr_data
## Dimensions: 320 x 1
## Duplicates: 183
##
## --------------------------------------------------------------------------------------------------------------
## No Variable Label Stats / Values Freqs (% of Valid) Graph Valid Missing
## ---- ----------- ------------- ---------------------- --------------------- ------------- ---------- ---------
## 1 Week_11 Week Eleven Mean (sd) : 8.5 (7) 137 distinct values : 320 0
## [numeric] min < med < max: : : (100.0%) (0.0%)
## 2 < 5.4 < 34 : :
## IQR (CV) : 5.4 (0.8) : :
## : : : : . .
## --------------------------------------------------------------------------------------------------------------dfSummary(cr_data$Week_13)## cr_data$Week_13 was converted to a data frame## Data Frame Summary
## cr_data
## Dimensions: 320 x 1
## Duplicates: 176
##
## -------------------------------------------------------------------------------------------------------------------
## No Variable Label Stats / Values Freqs (% of Valid) Graph Valid Missing
## ---- ----------- --------------- ----------------------- --------------------- --------------- ---------- ---------
## 1 Week_13 Week Thirteen Mean (sd) : 9.2 (8.4) 144 distinct values : . 320 0
## [numeric] min < med < max: : : (100.0%) (0.0%)
## 2 < 5.5 < 40 : :
## IQR (CV) : 5.6 (0.9) : :
## : : . : . . .
## -------------------------------------------------------------------------------------------------------------------Features by Behaviours
ggplot(transformed_boluwatife, aes(y =Week_7, col= as.factor(Behaviour)))+geom_boxplot()
ggplot(transformed_boluwatife, aes(y =Week_0, col= as.factor(Behaviour)))+geom_boxplot()
ggplot(transformed_boluwatife, aes(y =Week_5, col= as.factor(Behaviour)))+geom_boxplot()
ggplot(transformed_boluwatife, aes(y =Week_9, col= as.factor(Behaviour)))+geom_boxplot()
ggplot(transformed_boluwatife, aes(y =Week_11, col= as.factor(Behaviour)))+geom_boxplot()
ggplot(transformed_boluwatife, aes(y =Week_13, col= as.factor(Behaviour)))+geom_boxplot()
ggplot(transformed_boluwatife, aes(y =Week_3, col= as.factor(Behaviour)))+geom_boxplot()
Splitting the dataset intro Training set and test set 0.75 and 0.25
cont <- transformed_boluwatife %>% select(3:9)
scaled <- cont
cate <- transformed_boluwatife %>% select(!3:9)
b <- select(ucl_decoded, Behaviour)
transformed_boluwatife <- cbind(cate, scaled, b)
t <- transformed_boluwatife %>% select(n_behaviour = 10)
transformed_boluwatife <- transformed_boluwatife %>% select(!10)
transformed_boluwatife <- cbind(transformed_boluwatife, t) %>% select(!1)
library(caret)## Warning: package 'caret' was built under R version 4.3.3## Loading required package: lattice##
## Attaching package: 'caret'## The following object is masked from 'package:purrr':
##
## liftlibrary(caTools)
sample <- sample.split(transformed_boluwatife$Group, SplitRatio = 0.75)
Train <- subset(transformed_boluwatife, sample ==T)
Test <- subset(transformed_boluwatife, sample==F)Model creation
library(e1071)##
## Attaching package: 'e1071'## The following objects are masked from 'package:dlookr':
##
## kurtosis, skewnessModels on for Group based on One features
Removing the behaviour column
Test_g <- Test %>% select(!n_behaviour)
Train_g <- Train %>% select(!n_behaviour)Logistic regression
Logit <- glm(Group~., data = Train_g, family = binomial(link = "logit"))
summary(Logit)##
## Call:
## glm(formula = Group ~ ., family = binomial(link = "logit"), data = Train_g)
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.27006 0.22927 1.178 0.23883
## Week_0 0.14432 0.05578 2.587 0.00967 **
## Week_3 0.04661 0.05472 0.852 0.39437
## Week_5 0.05878 0.04495 1.308 0.19102
## Week_7 0.04148 0.04622 0.897 0.36948
## Week_9 -0.03144 0.04931 -0.637 0.52381
## Week_11 -0.13925 0.06398 -2.177 0.02951 *
## Week_13 -0.15076 0.05294 -2.848 0.00440 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 332.71 on 239 degrees of freedom
## Residual deviance: 303.24 on 232 degrees of freedom
## AIC: 319.24
##
## Number of Fisher Scoring iterations: 5pl <- predict(Logit, newdata = Test_g)
datpl <- as.factor(ifelse(pl >= 0.5, 1, 0))
confpl <- confusionMatrix(as.factor(Test_g$Group), datpl)
confpl## Confusion Matrix and Statistics
##
## Reference
## Prediction 0 1
## 0 37 3
## 1 37 3
##
## Accuracy : 0.5
## 95% CI : (0.386, 0.614)
## No Information Rate : 0.925
## P-Value [Acc > NIR] : 1
##
## Kappa : 0
##
## Mcnemar's Test P-Value : 1.811e-07
##
## Sensitivity : 0.5000
## Specificity : 0.5000
## Pos Pred Value : 0.9250
## Neg Pred Value : 0.0750
## Prevalence : 0.9250
## Detection Rate : 0.4625
## Detection Prevalence : 0.5000
## Balanced Accuracy : 0.5000
##
## 'Positive' Class : 0
## Probit regression
Probit <-glm(Group~., data = Train_g, family = binomial(link="probit"))
summary(Probit)##
## Call:
## glm(formula = Group ~ ., family = binomial(link = "probit"),
## data = Train_g)
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.17047 0.14080 1.211 0.2260
## Week_0 0.08276 0.03249 2.547 0.0109 *
## Week_3 0.02704 0.03142 0.861 0.3895
## Week_5 0.03402 0.02648 1.285 0.1988
## Week_7 0.02337 0.02671 0.875 0.3816
## Week_9 -0.01965 0.02914 -0.674 0.5001
## Week_11 -0.07847 0.03722 -2.108 0.0350 *
## Week_13 -0.08855 0.03059 -2.895 0.0038 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 332.71 on 239 degrees of freedom
## Residual deviance: 303.43 on 232 degrees of freedom
## AIC: 319.43
##
## Number of Fisher Scoring iterations: 5pb <- predict(Probit, newdata = Test_g)
datp <- as.factor(ifelse(pb >= 0.5, 1, 0))
confp <- confusionMatrix(as.factor(Test_g$Group), datp)
confp## Confusion Matrix and Statistics
##
## Reference
## Prediction 0 1
## 0 39 1
## 1 38 2
##
## Accuracy : 0.5125
## 95% CI : (0.3981, 0.6259)
## No Information Rate : 0.9625
## P-Value [Acc > NIR] : 1
##
## Kappa : 0.025
##
## Mcnemar's Test P-Value : 8.185e-09
##
## Sensitivity : 0.5065
## Specificity : 0.6667
## Pos Pred Value : 0.9750
## Neg Pred Value : 0.0500
## Prevalence : 0.9625
## Detection Rate : 0.4875
## Detection Prevalence : 0.5000
## Balanced Accuracy : 0.5866
##
## 'Positive' Class : 0
## Decision tree model
library(rpart)## Warning: package 'rpart' was built under R version 4.3.3library(rpart.plot)## Warning: package 'rpart.plot' was built under R version 4.3.3cart <- rpart(Group~., data = Train_g)
summary(cart)## Call:
## rpart(formula = Group ~ ., data = Train_g)
## n= 240
##
## CP nsplit rel error xerror xstd
## 1 0.22309905 0 1.0000000 1.0120542 0.003172568
## 2 0.08470146 1 0.7769010 0.9630955 0.056783151
## 3 0.07024795 2 0.6921995 0.9324862 0.068752465
## 4 0.06723783 3 0.6219515 0.9068161 0.071453267
## 5 0.03669618 4 0.5547137 0.9020017 0.076557643
## 6 0.02133343 5 0.5180175 0.8573562 0.079355754
## 7 0.01607143 6 0.4966841 0.9065573 0.084071604
## 8 0.01407625 8 0.4645412 0.9133495 0.085757790
## 9 0.01212121 9 0.4504650 0.9184050 0.085260447
## 10 0.01052291 10 0.4383438 0.9145756 0.085296731
## 11 0.01000000 11 0.4278209 0.9228764 0.086130320
##
## Variable importance
## Week_13 Week_5 Week_11 Week_7 Week_3 Week_9 Week_0
## 22 18 17 14 13 11 5
##
## Node number 1: 240 observations, complexity param=0.223099
## mean=0.5, MSE=0.25
## left son=2 (149 obs) right son=3 (91 obs)
## Primary splits:
## Week_13 < 5.015 to the right, improve=0.2230990, (0 missing)
## Week_11 < 4.255 to the right, improve=0.2072345, (0 missing)
## Week_5 < 4.41 to the right, improve=0.2063492, (0 missing)
## Week_9 < 3.92 to the right, improve=0.1927519, (0 missing)
## Week_7 < 5.01 to the right, improve=0.1785714, (0 missing)
## Surrogate splits:
## Week_5 < 4.92 to the right, agree=0.825, adj=0.538, (0 split)
## Week_9 < 4.895 to the right, agree=0.825, adj=0.538, (0 split)
## Week_11 < 4.255 to the right, agree=0.817, adj=0.516, (0 split)
## Week_7 < 5.06 to the right, agree=0.808, adj=0.495, (0 split)
## Week_3 < 4.245 to the right, agree=0.783, adj=0.429, (0 split)
##
## Node number 2: 149 observations, complexity param=0.08470146
## mean=0.3154362, MSE=0.2159362
## left son=4 (101 obs) right son=5 (48 obs)
## Primary splits:
## Week_3 < 10.5 to the left, improve=0.1579539, (0 missing)
## Week_5 < 10.5 to the left, improve=0.1544955, (0 missing)
## Week_0 < 7.8 to the left, improve=0.1517745, (0 missing)
## Week_7 < 10.5 to the left, improve=0.1517745, (0 missing)
## Week_11 < 10.5 to the left, improve=0.1353075, (0 missing)
## Surrogate splits:
## Week_0 < 7.8 to the left, agree=0.946, adj=0.833, (0 split)
## Week_11 < 9.65 to the left, agree=0.940, adj=0.812, (0 split)
## Week_13 < 9.7 to the left, agree=0.940, adj=0.812, (0 split)
## Week_5 < 11.5 to the left, agree=0.933, adj=0.792, (0 split)
## Week_7 < 9.31 to the left, agree=0.933, adj=0.792, (0 split)
##
## Node number 3: 91 observations, complexity param=0.07024795
## mean=0.8021978, MSE=0.1586765
## left son=6 (12 obs) right son=7 (79 obs)
## Primary splits:
## Week_5 < 5.2 to the right, improve=0.2918979, (0 missing)
## Week_9 < 5.32 to the right, improve=0.2429258, (0 missing)
## Week_11 < 5.835 to the right, improve=0.2326674, (0 missing)
## Week_7 < 4.815 to the right, improve=0.1901038, (0 missing)
## Week_13 < 3.95 to the right, improve=0.1506219, (0 missing)
## Surrogate splits:
## Week_11 < 6.75 to the right, agree=0.901, adj=0.250, (0 split)
## Week_9 < 5.865 to the right, agree=0.890, adj=0.167, (0 split)
##
## Node number 4: 101 observations, complexity param=0.06723783
## mean=0.1881188, MSE=0.1527301
## left son=8 (86 obs) right son=9 (15 obs)
## Primary splits:
## Week_11 < 4.25 to the right, improve=0.26152840, (0 missing)
## Week_3 < 4.235 to the right, improve=0.15725290, (0 missing)
## Week_7 < 4.95 to the right, improve=0.10527100, (0 missing)
## Week_13 < 5.75 to the right, improve=0.09828623, (0 missing)
## Week_5 < 4.36 to the right, improve=0.09329401, (0 missing)
## Surrogate splits:
## Week_5 < 3.935 to the right, agree=0.921, adj=0.467, (0 split)
## Week_9 < 4.055 to the right, agree=0.921, adj=0.467, (0 split)
## Week_7 < 4.1 to the right, agree=0.901, adj=0.333, (0 split)
## Week_3 < 3.605 to the right, agree=0.881, adj=0.200, (0 split)
##
## Node number 5: 48 observations, complexity param=0.03669618
## mean=0.5833333, MSE=0.2430556
## left son=10 (17 obs) right son=11 (31 obs)
## Primary splits:
## Week_13 < 22.5 to the right, improve=0.18872320, (0 missing)
## Week_9 < 16.5 to the right, improve=0.11454500, (0 missing)
## Week_11 < 22.5 to the right, improve=0.08691729, (0 missing)
## Week_7 < 24.5 to the left, improve=0.06938776, (0 missing)
## Week_5 < 13.5 to the left, improve=0.06222001, (0 missing)
## Surrogate splits:
## Week_7 < 12.5 to the left, agree=0.750, adj=0.294, (0 split)
## Week_9 < 10.5 to the left, agree=0.708, adj=0.176, (0 split)
## Week_0 < 29.5 to the right, agree=0.688, adj=0.118, (0 split)
## Week_11 < 21.5 to the right, agree=0.667, adj=0.059, (0 split)
##
## Node number 6: 12 observations
## mean=0.25, MSE=0.1875
##
## Node number 7: 79 observations, complexity param=0.02133343
## mean=0.8860759, MSE=0.1009454
## left son=14 (28 obs) right son=15 (51 obs)
## Primary splits:
## Week_13 < 3.95 to the right, improve=0.16050860, (0 missing)
## Week_9 < 4.96 to the right, improve=0.11749710, (0 missing)
## Week_11 < 5.33 to the right, improve=0.11749710, (0 missing)
## Week_7 < 4.51 to the right, improve=0.08785492, (0 missing)
## Week_0 < 5.65 to the right, improve=0.06131519, (0 missing)
## Surrogate splits:
## Week_5 < 4.25 to the right, agree=0.709, adj=0.179, (0 split)
## Week_9 < 4.005 to the right, agree=0.709, adj=0.179, (0 split)
## Week_11 < 4.89 to the right, agree=0.696, adj=0.143, (0 split)
## Week_0 < 6.15 to the right, agree=0.671, adj=0.071, (0 split)
## Week_3 < 4.96 to the right, agree=0.658, adj=0.036, (0 split)
##
## Node number 8: 86 observations, complexity param=0.01052291
## mean=0.1046512, MSE=0.0936993
## left son=16 (73 obs) right son=17 (13 obs)
## Primary splits:
## Week_11 < 4.7 to the right, improve=0.07835239, (0 missing)
## Week_7 < 4.95 to the right, improve=0.07603926, (0 missing)
## Week_0 < 4.05 to the right, improve=0.06523491, (0 missing)
## Week_3 < 4.235 to the right, improve=0.06523491, (0 missing)
## Week_13 < 5.65 to the right, improve=0.05197897, (0 missing)
## Surrogate splits:
## Week_5 < 4.46 to the right, agree=0.872, adj=0.154, (0 split)
## Week_0 < 3.73 to the right, agree=0.860, adj=0.077, (0 split)
## Week_9 < 4.105 to the right, agree=0.860, adj=0.077, (0 split)
##
## Node number 9: 15 observations
## mean=0.6666667, MSE=0.2222222
##
## Node number 10: 17 observations
## mean=0.2941176, MSE=0.2076125
##
## Node number 11: 31 observations, complexity param=0.01407625
## mean=0.7419355, MSE=0.1914672
## left son=22 (22 obs) right son=23 (9 obs)
## Primary splits:
## Week_7 < 17.5 to the right, improve=0.14229250, (0 missing)
## Week_9 < 17 to the right, improve=0.07617754, (0 missing)
## Week_11 < 14.5 to the left, improve=0.05010352, (0 missing)
## Week_3 < 17 to the right, improve=0.04614076, (0 missing)
## Week_5 < 27.5 to the left, improve=0.03216564, (0 missing)
## Surrogate splits:
## Week_9 < 14.5 to the right, agree=0.774, adj=0.222, (0 split)
## Week_0 < 24.5 to the left, agree=0.742, adj=0.111, (0 split)
## Week_3 < 28.5 to the left, agree=0.742, adj=0.111, (0 split)
##
## Node number 14: 28 observations, complexity param=0.01607143
## mean=0.7142857, MSE=0.2040816
## left son=28 (21 obs) right son=29 (7 obs)
## Primary splits:
## Week_3 < 3.505 to the right, improve=0.13333330, (0 missing)
## Week_7 < 4.815 to the right, improve=0.13333330, (0 missing)
## Week_11 < 5.23 to the right, improve=0.13333330, (0 missing)
## Week_0 < 3.93 to the left, improve=0.09037433, (0 missing)
## Week_13 < 4.15 to the left, improve=0.09037433, (0 missing)
## Surrogate splits:
## Week_0 < 2.74 to the right, agree=0.821, adj=0.286, (0 split)
## Week_5 < 3.305 to the right, agree=0.786, adj=0.143, (0 split)
## Week_11 < 3.59 to the right, agree=0.786, adj=0.143, (0 split)
##
## Node number 15: 51 observations
## mean=0.9803922, MSE=0.01922338
##
## Node number 16: 73 observations
## mean=0.06849315, MSE=0.06380184
##
## Node number 17: 13 observations
## mean=0.3076923, MSE=0.2130178
##
## Node number 22: 22 observations, complexity param=0.01212121
## mean=0.6363636, MSE=0.231405
## left son=44 (11 obs) right son=45 (11 obs)
## Primary splits:
## Week_7 < 25 to the left, improve=0.14285710, (0 missing)
## Week_9 < 19.5 to the right, improve=0.06696429, (0 missing)
## Week_0 < 18.5 to the right, improve=0.06696429, (0 missing)
## Week_5 < 17.5 to the right, improve=0.05982906, (0 missing)
## Week_11 < 14.5 to the left, improve=0.01953602, (0 missing)
## Surrogate splits:
## Week_3 < 19.5 to the left, agree=0.773, adj=0.545, (0 split)
## Week_0 < 12 to the right, agree=0.727, adj=0.455, (0 split)
## Week_5 < 18.5 to the right, agree=0.636, adj=0.273, (0 split)
## Week_9 < 26.5 to the left, agree=0.591, adj=0.182, (0 split)
## Week_11 < 15.5 to the right, agree=0.591, adj=0.182, (0 split)
##
## Node number 23: 9 observations
## mean=1, MSE=0
##
## Node number 28: 21 observations, complexity param=0.01607143
## mean=0.6190476, MSE=0.2358277
## left son=56 (7 obs) right son=57 (14 obs)
## Primary splits:
## Week_13 < 4.15 to the left, improve=0.23557690, (0 missing)
## Week_0 < 3.93 to the left, improve=0.15541790, (0 missing)
## Week_3 < 3.96 to the left, improve=0.07692308, (0 missing)
## Week_7 < 4.815 to the right, improve=0.07692308, (0 missing)
## Week_11 < 5.13 to the right, improve=0.07692308, (0 missing)
## Surrogate splits:
## Week_3 < 3.755 to the left, agree=0.762, adj=0.286, (0 split)
## Week_0 < 4.13 to the left, agree=0.714, adj=0.143, (0 split)
## Week_9 < 3.47 to the left, agree=0.714, adj=0.143, (0 split)
##
## Node number 29: 7 observations
## mean=1, MSE=0
##
## Node number 44: 11 observations
## mean=0.4545455, MSE=0.2479339
##
## Node number 45: 11 observations
## mean=0.8181818, MSE=0.1487603
##
## Node number 56: 7 observations
## mean=0.2857143, MSE=0.2040816
##
## Node number 57: 14 observations
## mean=0.7857143, MSE=0.1683673rpart.plot(cart)
pre <- predict(cart, newdata = Test_g)
dat_pre <- as.factor(ifelse(pre >= 0.5, 1, 0))
conf_matr <- confusionMatrix(as.factor(Test_g$Group), dat_pre)
conf_matr## Confusion Matrix and Statistics
##
## Reference
## Prediction 0 1
## 0 30 10
## 1 15 25
##
## Accuracy : 0.6875
## 95% CI : (0.5741, 0.7865)
## No Information Rate : 0.5625
## P-Value [Acc > NIR] : 0.01511
##
## Kappa : 0.375
##
## Mcnemar's Test P-Value : 0.42371
##
## Sensitivity : 0.6667
## Specificity : 0.7143
## Pos Pred Value : 0.7500
## Neg Pred Value : 0.6250
## Prevalence : 0.5625
## Detection Rate : 0.3750
## Detection Prevalence : 0.5000
## Balanced Accuracy : 0.6905
##
## 'Positive' Class : 0
## Random Forest
library(randomForest)## Warning: package 'randomForest' was built under R version 4.3.3## randomForest 4.7-1.1## Type rfNews() to see new features/changes/bug fixes.##
## Attaching package: 'randomForest'## The following object is masked from 'package:dplyr':
##
## combine## The following object is masked from 'package:ggplot2':
##
## marginl <- randomForest(Group~., data = Train_g)## Warning in randomForest.default(m, y, ...): The response has five or fewer
## unique values. Are you sure you want to do regression?summary(l)## Length Class Mode
## call 3 -none- call
## type 1 -none- character
## predicted 240 -none- numeric
## mse 500 -none- numeric
## rsq 500 -none- numeric
## oob.times 240 -none- numeric
## importance 7 -none- numeric
## importanceSD 0 -none- NULL
## localImportance 0 -none- NULL
## proximity 0 -none- NULL
## ntree 1 -none- numeric
## mtry 1 -none- numeric
## forest 11 -none- list
## coefs 0 -none- NULL
## y 240 -none- numeric
## test 0 -none- NULL
## inbag 0 -none- NULL
## terms 3 terms callplot(l)
importance(l)## IncNodePurity
## Week_0 4.872464
## Week_3 6.103293
## Week_5 8.532478
## Week_7 7.867519
## Week_9 7.709972
## Week_11 9.523452
## Week_13 10.510064library(randomForestExplainer)## Warning: package 'randomForestExplainer' was built under R version 4.3.3## Registered S3 method overwritten by 'GGally':
## method from
## +.gg ggplot2tree_to_plot <- getTree(l, k = 1, labelVar = TRUE)
plot(tree_to_plot)
predi <- predict(l, newdata = Test_g) dat_pred <-
as.factor(ifelse(predi >= 0.5, 1, 0)) conf_matri <-
confusionMatrix(as.factor(Test_g$Group), dat_pred) conf_matri
Support Vector Machine
library(e1071)
suport <- svm(Group~., data = Train_g, cost = 0.1, type="C-classification", kernel ="linear", scale = T)
summary(suport)##
## Call:
## svm(formula = Group ~ ., data = Train_g, cost = 0.1, type = "C-classification",
## kernel = "linear", scale = T)
##
##
## Parameters:
## SVM-Type: C-classification
## SVM-Kernel: linear
## cost: 0.1
##
## Number of Support Vectors: 219
##
## ( 109 110 )
##
##
## Number of Classes: 2
##
## Levels:
## 0 1Extract two principal components for visualization
pca_data <- prcomp(Train_g[, -ncol(Train_g)], scale. = TRUE)
data_pca <- as.data.frame(cbind(pca_data$x[,1], pca_data$x[,2], Group = Train_g$Group))Plot decision boundary
ggplot(data_pca, aes(x = V1, y = V2, color = Group)) +
geom_point() +
labs(x = "Principal Component 1", y = "Principal Component 2", color = "Label") +
theme_minimal()
Hyperparameter tuning
tuned <- tune(svm, Group~., data = Train_g, kernel= "linear", ranges = list(cost =c(0.1,1, 10, 20, 100)))Print the results
print(tuned) ##
## Parameter tuning of 'svm':
##
## - sampling method: 10-fold cross validation
##
## - best parameters:
## cost
## 0.1
##
## - best performance: 0.3288493pred <- predict(suport, Test_g)Compute confusion matrix and other evaluation parameters
conf_matrix <- confusionMatrix(as.factor(Test_g$Group), pred)
conf_matrix## Confusion Matrix and Statistics
##
## Reference
## Prediction 0 1
## 0 10 30
## 1 7 33
##
## Accuracy : 0.5375
## 95% CI : (0.4224, 0.6497)
## No Information Rate : 0.7875
## P-Value [Acc > NIR] : 0.9999998
##
## Kappa : 0.075
##
## Mcnemar's Test P-Value : 0.0002983
##
## Sensitivity : 0.5882
## Specificity : 0.5238
## Pos Pred Value : 0.2500
## Neg Pred Value : 0.8250
## Prevalence : 0.2125
## Detection Rate : 0.1250
## Detection Prevalence : 0.5000
## Balanced Accuracy : 0.5560
##
## 'Positive' Class : 0
## MODEL FOR BEHAVIOUR
Classification and Regression Tree
Test_b <- transformed_boluwatife %>% select(-Group)
Train_b <- transformed_boluwatife %>% select(-Group)
cart_behaviour <- rpart(as.factor(n_behaviour)~., data = Train_b)
summary(cart_behaviour)## Call:
## rpart(formula = as.factor(n_behaviour) ~ ., data = Train_b)
## n= 320
##
## CP nsplit rel error xerror xstd
## 1 0.33333333 0 1.0000000 1.1250000 0.02706329
## 2 0.15833333 1 0.6666667 0.7708333 0.03681006
## 3 0.03333333 2 0.5083333 0.6000000 0.03708099
## 4 0.02916667 3 0.4750000 0.6166667 0.03716284
## 5 0.02083333 4 0.4458333 0.5541667 0.03673334
## 6 0.01944444 6 0.4041667 0.5166667 0.03631221
## 7 0.01666667 9 0.3458333 0.5083333 0.03620148
## 8 0.01000000 10 0.3291667 0.5208333 0.03636521
##
## Variable importance
## Week_11 Week_0 Week_5 Week_13 Week_7 Week_9 Week_3
## 17 17 16 16 16 16 3
##
## Node number 1: 320 observations, complexity param=0.3333333
## predicted class= TL expected loss=0.75 P(node) =1
## class counts: 80 80 80 80
## probabilities: 0.250 0.250 0.250 0.250
## left son=2 (80 obs) right son=3 (240 obs)
## Primary splits:
## Week_11 < 9.65 to the right, improve=80.00000, (0 missing)
## Week_13 < 9.7 to the right, improve=80.00000, (0 missing)
## Week_9 < 9.6 to the right, improve=78.67220, (0 missing)
## Week_7 < 9.55 to the right, improve=77.35537, (0 missing)
## Week_5 < 9.5 to the right, improve=74.75410, (0 missing)
## Surrogate splits:
## Week_13 < 9.7 to the right, agree=1.000, adj=1.000, (0 split)
## Week_9 < 9.6 to the right, agree=0.997, adj=0.987, (0 split)
## Week_7 < 9.55 to the right, agree=0.994, adj=0.975, (0 split)
## Week_5 < 9.5 to the right, agree=0.988, adj=0.950, (0 split)
## Week_0 < 7.995 to the right, agree=0.984, adj=0.938, (0 split)
##
## Node number 2: 80 observations
## predicted class=Aggressive expected loss=0 P(node) =0.25
## class counts: 0 80 0 0
## probabilities: 0.000 1.000 0.000 0.000
##
## Node number 3: 240 observations, complexity param=0.1583333
## predicted class= TL expected loss=0.6666667 P(node) =0.75
## class counts: 80 0 80 80
## probabilities: 0.333 0.000 0.333 0.333
## left son=6 (164 obs) right son=7 (76 obs)
## Primary splits:
## Week_3 < 3.995 to the right, improve=14.22336, (0 missing)
## Week_0 < 4.895 to the right, improve=13.49114, (0 missing)
## Week_13 < 5.58 to the right, improve=13.09218, (0 missing)
## Week_11 < 5.45 to the right, improve=13.02944, (0 missing)
## Week_9 < 5.375 to the right, improve=12.91925, (0 missing)
## Surrogate splits:
## Week_5 < 3.92 to the right, agree=0.775, adj=0.289, (0 split)
## Week_0 < 3.91 to the right, agree=0.762, adj=0.250, (0 split)
## Week_11 < 3.795 to the right, agree=0.750, adj=0.211, (0 split)
## Week_13 < 3.975 to the right, agree=0.746, adj=0.197, (0 split)
## Week_9 < 3.49 to the right, agree=0.733, adj=0.158, (0 split)
##
## Node number 6: 164 observations, complexity param=0.03333333
## predicted class= TL expected loss=0.5609756 P(node) =0.5125
## class counts: 72 0 58 34
## probabilities: 0.439 0.000 0.354 0.207
## left son=12 (74 obs) right son=13 (90 obs)
## Primary splits:
## Week_7 < 5.28 to the right, improve=6.251491, (0 missing)
## Week_0 < 4.895 to the right, improve=5.776332, (0 missing)
## Week_9 < 5.37 to the right, improve=5.604068, (0 missing)
## Week_13 < 5.58 to the right, improve=5.604068, (0 missing)
## Week_5 < 3.99 to the right, improve=5.348671, (0 missing)
## Surrogate splits:
## Week_13 < 5.84 to the right, agree=0.720, adj=0.378, (0 split)
## Week_9 < 5.85 to the right, agree=0.695, adj=0.324, (0 split)
## Week_11 < 5.885 to the right, agree=0.683, adj=0.297, (0 split)
## Week_3 < 5.995 to the right, agree=0.677, adj=0.284, (0 split)
## Week_5 < 5.15 to the right, agree=0.659, adj=0.243, (0 split)
##
## Node number 7: 76 observations, complexity param=0.02916667
## predicted class=Weight expected loss=0.3947368 P(node) =0.2375
## class counts: 8 0 22 46
## probabilities: 0.105 0.000 0.289 0.605
## left son=14 (12 obs) right son=15 (64 obs)
## Primary splits:
## Week_0 < 5.015 to the right, improve=6.207785, (0 missing)
## Week_11 < 4.89 to the right, improve=5.108541, (0 missing)
## Week_7 < 4.195 to the right, improve=3.583801, (0 missing)
## Week_13 < 5.49 to the right, improve=3.050888, (0 missing)
## Week_9 < 4.075 to the right, improve=2.581998, (0 missing)
## Surrogate splits:
## Week_13 < 5.49 to the right, agree=0.882, adj=0.250, (0 split)
## Week_9 < 5.445 to the right, agree=0.868, adj=0.167, (0 split)
## Week_7 < 5.35 to the right, agree=0.855, adj=0.083, (0 split)
## Week_11 < 5.3 to the right, agree=0.855, adj=0.083, (0 split)
##
## Node number 12: 74 observations, complexity param=0.01666667
## predicted class= TL expected loss=0.3918919 P(node) =0.23125
## class counts: 45 0 23 6
## probabilities: 0.608 0.000 0.311 0.081
## left son=24 (56 obs) right son=25 (18 obs)
## Primary splits:
## Week_0 < 4.85 to the right, improve=3.242063, (0 missing)
## Week_9 < 5.35 to the right, improve=2.710008, (0 missing)
## Week_11 < 5.45 to the right, improve=2.704215, (0 missing)
## Week_5 < 5.05 to the right, improve=1.860140, (0 missing)
## Week_13 < 5.55 to the right, improve=1.754286, (0 missing)
## Surrogate splits:
## Week_5 < 3.2 to the right, agree=0.784, adj=0.111, (0 split)
## Week_3 < 4.335 to the right, agree=0.770, adj=0.056, (0 split)
## Week_13 < 3.75 to the right, agree=0.770, adj=0.056, (0 split)
##
## Node number 13: 90 observations, complexity param=0.02083333
## predicted class=SL expected loss=0.6111111 P(node) =0.28125
## class counts: 27 0 35 28
## probabilities: 0.300 0.000 0.389 0.311
## left son=26 (70 obs) right son=27 (20 obs)
## Primary splits:
## Week_0 < 3.995 to the right, improve=3.549206, (0 missing)
## Week_5 < 3.99 to the right, improve=3.452109, (0 missing)
## Week_13 < 3.73 to the right, improve=1.628453, (0 missing)
## Week_11 < 5.575 to the right, improve=1.366548, (0 missing)
## Week_9 < 3.625 to the left, improve=1.023724, (0 missing)
## Surrogate splits:
## Week_9 < 2.65 to the right, agree=0.789, adj=0.05, (0 split)
##
## Node number 14: 12 observations
## predicted class=SL expected loss=0.3333333 P(node) =0.0375
## class counts: 3 0 8 1
## probabilities: 0.250 0.000 0.667 0.083
##
## Node number 15: 64 observations
## predicted class=Weight expected loss=0.296875 P(node) =0.2
## class counts: 5 0 14 45
## probabilities: 0.078 0.000 0.219 0.703
##
## Node number 24: 56 observations
## predicted class= TL expected loss=0.3035714 P(node) =0.175
## class counts: 39 0 13 4
## probabilities: 0.696 0.000 0.232 0.071
##
## Node number 25: 18 observations
## predicted class=SL expected loss=0.4444444 P(node) =0.05625
## class counts: 6 0 10 2
## probabilities: 0.333 0.000 0.556 0.111
##
## Node number 26: 70 observations, complexity param=0.01944444
## predicted class= TL expected loss=0.6142857 P(node) =0.21875
## class counts: 27 0 25 18
## probabilities: 0.386 0.000 0.357 0.257
## left son=52 (36 obs) right son=53 (34 obs)
## Primary splits:
## Week_11 < 4.865 to the left, improve=1.6658260, (0 missing)
## Week_3 < 5.15 to the left, improve=1.3785710, (0 missing)
## Week_7 < 4.915 to the left, improve=1.0285710, (0 missing)
## Week_5 < 6.25 to the left, improve=1.0231070, (0 missing)
## Week_0 < 6.25 to the left, improve=0.9237327, (0 missing)
## Surrogate splits:
## Week_7 < 4.21 to the left, agree=0.771, adj=0.529, (0 split)
## Week_5 < 4.96 to the left, agree=0.729, adj=0.441, (0 split)
## Week_9 < 4.975 to the left, agree=0.686, adj=0.353, (0 split)
## Week_13 < 4.75 to the left, agree=0.686, adj=0.353, (0 split)
## Week_0 < 5.15 to the left, agree=0.614, adj=0.206, (0 split)
##
## Node number 27: 20 observations, complexity param=0.02083333
## predicted class=SL expected loss=0.5 P(node) =0.0625
## class counts: 0 0 10 10
## probabilities: 0.000 0.000 0.500 0.500
## left son=54 (12 obs) right son=55 (8 obs)
## Primary splits:
## Week_5 < 3.99 to the right, improve=6.6666670, (0 missing)
## Week_13 < 4.875 to the right, improve=2.7472530, (0 missing)
## Week_11 < 4.255 to the right, improve=2.5252530, (0 missing)
## Week_7 < 4.025 to the right, improve=1.6000000, (0 missing)
## Week_9 < 4.21 to the left, improve=0.4166667, (0 missing)
## Surrogate splits:
## Week_11 < 4.255 to the right, agree=0.75, adj=0.375, (0 split)
## Week_0 < 3.315 to the right, agree=0.70, adj=0.250, (0 split)
## Week_7 < 2.93 to the right, agree=0.70, adj=0.250, (0 split)
## Week_9 < 4.495 to the right, agree=0.65, adj=0.125, (0 split)
## Week_13 < 4.875 to the right, agree=0.65, adj=0.125, (0 split)
##
## Node number 52: 36 observations, complexity param=0.01944444
## predicted class= TL expected loss=0.5277778 P(node) =0.1125
## class counts: 17 0 14 5
## probabilities: 0.472 0.000 0.389 0.139
## left son=104 (17 obs) right son=105 (19 obs)
## Primary splits:
## Week_11 < 3.975 to the right, improve=2.724974, (0 missing)
## Week_9 < 3.925 to the right, improve=1.776190, (0 missing)
## Week_3 < 5.505 to the left, improve=1.462963, (0 missing)
## Week_7 < 3.245 to the right, improve=1.389984, (0 missing)
## Week_5 < 3.95 to the right, improve=1.333333, (0 missing)
## Surrogate splits:
## Week_9 < 4.025 to the right, agree=0.694, adj=0.353, (0 split)
## Week_13 < 3.85 to the right, agree=0.694, adj=0.353, (0 split)
## Week_0 < 5.95 to the right, agree=0.639, adj=0.235, (0 split)
## Week_5 < 3.76 to the right, agree=0.639, adj=0.235, (0 split)
## Week_7 < 4.065 to the right, agree=0.639, adj=0.235, (0 split)
##
## Node number 53: 34 observations, complexity param=0.01944444
## predicted class=Weight expected loss=0.6176471 P(node) =0.10625
## class counts: 10 0 11 13
## probabilities: 0.294 0.000 0.324 0.382
## left son=106 (20 obs) right son=107 (14 obs)
## Primary splits:
## Week_11 < 5.575 to the right, improve=3.943697, (0 missing)
## Week_13 < 5.58 to the right, improve=3.472269, (0 missing)
## Week_0 < 4.3 to the left, improve=2.561158, (0 missing)
## Week_9 < 5.57 to the right, improve=2.200840, (0 missing)
## Week_5 < 6.25 to the right, improve=1.683258, (0 missing)
## Surrogate splits:
## Week_5 < 4.86 to the right, agree=0.735, adj=0.357, (0 split)
## Week_13 < 4.625 to the right, agree=0.706, adj=0.286, (0 split)
## Week_7 < 3.97 to the right, agree=0.676, adj=0.214, (0 split)
## Week_9 < 3.84 to the right, agree=0.676, adj=0.214, (0 split)
## Week_3 < 6.3 to the left, agree=0.618, adj=0.071, (0 split)
##
## Node number 54: 12 observations
## predicted class=SL expected loss=0.1666667 P(node) =0.0375
## class counts: 0 0 10 2
## probabilities: 0.000 0.000 0.833 0.167
##
## Node number 55: 8 observations
## predicted class=Weight expected loss=0 P(node) =0.025
## class counts: 0 0 0 8
## probabilities: 0.000 0.000 0.000 1.000
##
## Node number 104: 17 observations
## predicted class= TL expected loss=0.2941176 P(node) =0.053125
## class counts: 12 0 4 1
## probabilities: 0.706 0.000 0.235 0.059
##
## Node number 105: 19 observations
## predicted class=SL expected loss=0.4736842 P(node) =0.059375
## class counts: 5 0 10 4
## probabilities: 0.263 0.000 0.526 0.211
##
## Node number 106: 20 observations
## predicted class=SL expected loss=0.55 P(node) =0.0625
## class counts: 8 0 9 3
## probabilities: 0.400 0.000 0.450 0.150
##
## Node number 107: 14 observations
## predicted class=Weight expected loss=0.2857143 P(node) =0.04375
## class counts: 2 0 2 10
## probabilities: 0.143 0.000 0.143 0.714rpart.plot(cart_behaviour)
pre_behaviour <- predict(cart_behaviour, newdata = Test_b, type ="class")
conf_matr_behaviour <- confusionMatrix(as.factor(Test_b$n_behaviour), pre_behaviour)
conf_matr_behaviour## Confusion Matrix and Statistics
##
## Reference
## Prediction TL Aggressive SL Weight
## TL 51 0 22 7
## Aggressive 0 80 0 0
## SL 17 0 47 16
## Weight 5 0 12 63
##
## Overall Statistics
##
## Accuracy : 0.7531
## 95% CI : (0.7021, 0.7994)
## No Information Rate : 0.2688
## P-Value [Acc > NIR] : < 2.2e-16
##
## Kappa : 0.6708
##
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: TL Class: Aggressive Class: SL Class: Weight
## Sensitivity 0.6986 1.00 0.5802 0.7326
## Specificity 0.8826 1.00 0.8619 0.9274
## Pos Pred Value 0.6375 1.00 0.5875 0.7875
## Neg Pred Value 0.9083 1.00 0.8583 0.9042
## Prevalence 0.2281 0.25 0.2531 0.2687
## Detection Rate 0.1594 0.25 0.1469 0.1969
## Detection Prevalence 0.2500 0.25 0.2500 0.2500
## Balanced Accuracy 0.7906 1.00 0.7211 0.8300Iterative dichotomiser 3
library(C50)## Warning: package 'C50' was built under R version 4.3.3ID_3 <- C5.0(as.factor(n_behaviour)~., data = Train_b)
summary(ID_3)##
## Call:
## C5.0.formula(formula = as.factor(n_behaviour) ~ ., data = Train_b)
##
##
## C5.0 [Release 2.07 GPL Edition] Wed Apr 17 21:46:09 2024
## -------------------------------
##
## Class specified by attribute `outcome'
##
## Read 320 cases (8 attributes) from undefined.data
##
## Decision tree:
##
## Week_11 > 9.3: Aggressive (80)
## Week_11 <= 9.3:
## :...Week_3 <= 3.99:
## :...Week_0 > 5:
## : :...Week_5 <= 3.5: TL (2/1)
## : : Week_5 > 3.5:
## : : :...Week_0 <= 5.8: SL (7)
## : : Week_0 > 5.8: TL (3/1)
## : Week_0 <= 5:
## : :...Week_7 > 4.19:
## : :...Week_5 <= 3.01: Weight (4)
## : : Week_5 > 3.01:
## : : :...Week_5 > 4:
## : : :...Week_3 <= 3.8: Weight (4)
## : : : Week_3 > 3.8: SL (3/1)
## : : Week_5 <= 4:
## : : :...Week_5 <= 3.21: SL (2)
## : : Week_5 > 3.21:
## : : :...Week_7 <= 4.82: TL (3)
## : : Week_7 > 4.82: SL (3/1)
## : Week_7 <= 4.19:
## : :...Week_13 > 3.6: Weight (25/1)
## : Week_13 <= 3.6:
## : :...Week_13 > 3.46: SL (3)
## : Week_13 <= 3.46:
## : :...Week_9 > 4.29: SL (3/1)
## : Week_9 <= 4.29:
## : :...Week_0 <= 4.1: Weight (10)
## : Week_0 > 4.1:
## : :...Week_5 <= 3.6: Weight (2)
## : Week_5 > 3.6: SL (2)
## Week_3 > 3.99:
## :...Week_0 <= 3.99:
## :...Week_5 <= 3.98: Weight (9/1)
## : Week_5 > 3.98:
## : :...Week_7 <= 5.23: SL (11/1)
## : Week_7 > 5.23: Weight (3/1)
## Week_0 > 3.99:
## :...Week_9 <= 5.35:
## :...Week_3 > 7.1: Weight (3)
## : Week_3 <= 7.1:
## : :...Week_5 <= 7.5: TL (71/43)
## : Week_5 > 7.5: Weight (2)
## Week_9 > 5.35:
## :...Week_11 > 7.43: TL (12)
## Week_11 <= 7.43:
## :...Week_7 > 5.9:
## :...Week_0 <= 4.4: SL (2)
## : Week_0 > 4.4:
## : :...Week_11 > 5.1: TL (18/1)
## : Week_11 <= 5.1:
## : :...Week_11 <= 4.8: TL (2)
## : Week_11 > 4.8: SL (3)
## Week_7 <= 5.9:
## :...Week_11 > 6.7: SL (6)
## Week_11 <= 6.7:
## :...Week_5 > 5.9:
## :...Week_9 <= 6.4: TL (5)
## : Week_9 > 6.4: Weight (2)
## Week_5 <= 5.9:
## :...Week_7 <= 3.95: TL (3/1)
## Week_7 > 3.95:
## :...Week_9 > 7.5: TL (2)
## Week_9 <= 7.5:
## :...Week_5 <= 5.6: SL (7)
## Week_5 > 5.6: TL (3/1)
##
##
## Evaluation on training data (320 cases):
##
## Decision Tree
## ----------------
## Size Errors
##
## 34 55(17.2%) <<
##
##
## (a) (b) (c) (d) <-classified as
## ---- ---- ---- ----
## 76 2 2 (a): class TL
## 80 (b): class Aggressive
## 31 48 1 (c): class SL
## 17 2 61 (d): class Weight
##
##
## Attribute usage:
##
## 100.00% Week_11
## 75.00% Week_0
## 75.00% Week_3
## 49.38% Week_9
## 47.81% Week_5
## 40.94% Week_7
## 14.06% Week_13
##
##
## Time: 0.0 secsplot(ID_3)
ID_3_behaviour <- predict(ID_3, newdata = Test_b, prob ="class")
ID_3_behaviour <- confusionMatrix(as.factor(Test_b$n_behaviour), ID_3_behaviour)
ID_3_behaviour## Confusion Matrix and Statistics
##
## Reference
## Prediction TL Aggressive SL Weight
## TL 76 0 2 2
## Aggressive 0 80 0 0
## SL 31 0 48 1
## Weight 17 0 2 61
##
## Overall Statistics
##
## Accuracy : 0.8281
## 95% CI : (0.7822, 0.8678)
## No Information Rate : 0.3875
## P-Value [Acc > NIR] : < 2.2e-16
##
## Kappa : 0.7708
##
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: TL Class: Aggressive Class: SL Class: Weight
## Sensitivity 0.6129 1.00 0.9231 0.9531
## Specificity 0.9796 1.00 0.8806 0.9258
## Pos Pred Value 0.9500 1.00 0.6000 0.7625
## Neg Pred Value 0.8000 1.00 0.9833 0.9875
## Prevalence 0.3875 0.25 0.1625 0.2000
## Detection Rate 0.2375 0.25 0.1500 0.1906
## Detection Prevalence 0.2500 0.25 0.2500 0.2500
## Balanced Accuracy 0.7962 1.00 0.9018 0.9395Chi-square automatic iterative detection Model
library(partykit)## Warning: package 'partykit' was built under R version 4.3.3## Loading required package: grid## Loading required package: libcoin## Warning: package 'libcoin' was built under R version 4.3.3## Loading required package: mvtnorm## Warning: package 'mvtnorm' was built under R version 4.3.3Chaid <- ctree(as.factor(n_behaviour) ~., data = Train_b)
summary(Chaid)## Length Class Mode
## 1 9 constparty list
## 2 7 constparty list
## 3 3 constparty list
## 4 1 constparty list
## 5 1 constparty list
## 6 3 constparty list
## 7 1 constparty list
## 8 1 constparty list
## 9 1 constparty listplot(Chaid)
chaid_behaviour <- predict(Chaid, newdata = Test_b, prob = "class")
## Compute confusion matrix
conf_matrix_chaid <- confusionMatrix(as.factor(Test_b$n_behaviour), chaid_behaviour)
conf_matrix_chaid## Confusion Matrix and Statistics
##
## Reference
## Prediction TL Aggressive SL Weight
## TL 57 0 18 5
## Aggressive 0 80 0 0
## SL 35 0 32 13
## Weight 17 0 19 44
##
## Overall Statistics
##
## Accuracy : 0.6656
## 95% CI : (0.611, 0.7171)
## No Information Rate : 0.3406
## P-Value [Acc > NIR] : < 2.2e-16
##
## Kappa : 0.5542
##
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: TL Class: Aggressive Class: SL Class: Weight
## Sensitivity 0.5229 1.00 0.4638 0.7097
## Specificity 0.8910 1.00 0.8088 0.8605
## Pos Pred Value 0.7125 1.00 0.4000 0.5500
## Neg Pred Value 0.7833 1.00 0.8458 0.9250
## Prevalence 0.3406 0.25 0.2156 0.1938
## Detection Rate 0.1781 0.25 0.1000 0.1375
## Detection Prevalence 0.2500 0.25 0.2500 0.2500
## Balanced Accuracy 0.7070 1.00 0.6363 0.7851## Conditional inference trees
library(party)## Warning: package 'party' was built under R version 4.3.3## Loading required package: modeltools## Loading required package: stats4## Loading required package: strucchange## Warning: package 'strucchange' was built under R version 4.3.3## Loading required package: zoo##
## Attaching package: 'zoo'## The following objects are masked from 'package:base':
##
## as.Date, as.Date.numeric## Loading required package: sandwich## Warning: package 'sandwich' was built under R version 4.3.3##
## Attaching package: 'strucchange'## The following object is masked from 'package:stringr':
##
## boundary##
## Attaching package: 'party'## The following objects are masked from 'package:partykit':
##
## cforest, ctree, ctree_control, edge_simple, mob, mob_control,
## node_barplot, node_bivplot, node_boxplot, node_inner, node_surv,
## node_terminal, varimp## The following object is masked from 'package:dplyr':
##
## wherel_behaviour <- ctree(as.factor(n_behaviour)~., data = Train_b)
summary(l_behaviour)## Length Class Mode
## 1 BinaryTree S4plot(l_behaviour)
predi_behaviour <- predict(l_behaviour, newdata = Test_b, prob ="class")
conf_matri_behavi <- confusionMatrix(as.factor(Test_b$n_behaviour), predi_behaviour)
conf_matri_behavi## Confusion Matrix and Statistics
##
## Reference
## Prediction TL Aggressive SL Weight
## TL 38 0 37 5
## Aggressive 0 80 0 0
## SL 23 0 38 19
## Weight 4 0 36 40
##
## Overall Statistics
##
## Accuracy : 0.6125
## 95% CI : (0.5567, 0.6662)
## No Information Rate : 0.3469
## P-Value [Acc > NIR] : < 2.2e-16
##
## Kappa : 0.4833
##
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: TL Class: Aggressive Class: SL Class: Weight
## Sensitivity 0.5846 1.00 0.3423 0.6250
## Specificity 0.8353 1.00 0.7990 0.8438
## Pos Pred Value 0.4750 1.00 0.4750 0.5000
## Neg Pred Value 0.8875 1.00 0.6958 0.9000
## Prevalence 0.2031 0.25 0.3469 0.2000
## Detection Rate 0.1187 0.25 0.1187 0.1250
## Detection Prevalence 0.2500 0.25 0.2500 0.2500
## Balanced Accuracy 0.7100 1.00 0.5707 0.7344Naive Bayes
behaviour <- naiveBayes(as.factor(n_behaviour)~., data = Train_b)
summary(behaviour)## Length Class Mode
## apriori 4 table numeric
## tables 7 -none- list
## levels 4 -none- character
## isnumeric 7 -none- logical
## call 4 -none- callprei_behaviour <- predict(behaviour, newdata = Test_b, type ="class")
conf_matri_behavi <- confusionMatrix(as.factor(Test_b$n_behaviour), as.factor(prei_behaviour))
conf_matri_behavi## Confusion Matrix and Statistics
##
## Reference
## Prediction TL Aggressive SL Weight
## TL 43 0 19 18
## Aggressive 0 80 0 0
## SL 27 0 18 35
## Weight 13 0 13 54
##
## Overall Statistics
##
## Accuracy : 0.6094
## 95% CI : (0.5535, 0.6632)
## No Information Rate : 0.3344
## P-Value [Acc > NIR] : < 2.2e-16
##
## Kappa : 0.4792
##
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: TL Class: Aggressive Class: SL Class: Weight
## Sensitivity 0.5181 1.00 0.36000 0.5047
## Specificity 0.8439 1.00 0.77037 0.8779
## Pos Pred Value 0.5375 1.00 0.22500 0.6750
## Neg Pred Value 0.8333 1.00 0.86667 0.7792
## Prevalence 0.2594 0.25 0.15625 0.3344
## Detection Rate 0.1344 0.25 0.05625 0.1688
## Detection Prevalence 0.2500 0.25 0.25000 0.2500
## Balanced Accuracy 0.6810 1.00 0.56519 0.6913Support Vector Machine
suport_behaviour <- svm(as.factor(n_behaviour)~., data = Train_b, kernel ="linear", cost = 10)
summary(suport_behaviour)##
## Call:
## svm(formula = as.factor(n_behaviour) ~ ., data = Train_b, kernel = "linear",
## cost = 10)
##
##
## Parameters:
## SVM-Type: C-classification
## SVM-Kernel: linear
## cost: 10
##
## Number of Support Vectors: 214
##
## ( 80 66 4 64 )
##
##
## Number of Classes: 4
##
## Levels:
## TL Aggressive SL Weighttuned <- tune(svm, as.factor(n_behaviour)~., data = Train_b, kernel= "linear", ranges = list(cost =c(0.1,1, 10, 20, 100)))
print(tuned) ##
## Parameter tuning of 'svm':
##
## - sampling method: 10-fold cross validation
##
## - best parameters:
## cost
## 1
##
## - best performance: 0.415625predq_behaviour <- predict(suport_behaviour, Test_b)
conf_matrix_behaviour <- confusionMatrix(as.factor(Test_b$n_behaviour), predq_behaviour)
conf_matrix_behaviour## Confusion Matrix and Statistics
##
## Reference
## Prediction TL Aggressive SL Weight
## TL 44 0 19 17
## Aggressive 0 80 0 0
## SL 25 0 24 31
## Weight 17 0 10 53
##
## Overall Statistics
##
## Accuracy : 0.6281
## 95% CI : (0.5726, 0.6812)
## No Information Rate : 0.3156
## P-Value [Acc > NIR] : < 2.2e-16
##
## Kappa : 0.5042
##
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: TL Class: Aggressive Class: SL Class: Weight
## Sensitivity 0.5116 1.00 0.4528 0.5248
## Specificity 0.8462 1.00 0.7903 0.8767
## Pos Pred Value 0.5500 1.00 0.3000 0.6625
## Neg Pred Value 0.8250 1.00 0.8792 0.8000
## Prevalence 0.2687 0.25 0.1656 0.3156
## Detection Rate 0.1375 0.25 0.0750 0.1656
## Detection Prevalence 0.2500 0.25 0.2500 0.2500
## Balanced Accuracy 0.6789 1.00 0.6215 0.7007Summary
This summary of model performance metrics provides insights into the accuracy, confidence intervals, agreement beyond chance (Kappa), and statistical significance of different predictive models. Here’s what each aspect means: Accuracy: It represents the proportion of correctly classified instances by the model. Higher accuracy indicates better predictive performance. 95% Confidence Intervals: They provide a range within which the true accuracy of the model is likely to fall with 95% confidence. Narrower intervals indicate more precise estimates. No Information Rate: It serves as a baseline accuracy level that a model should outperform. It represents the accuracy achieved by predicting the majority class in the dataset. Kappa: It measures the agreement between predicted and observed classifications, accounting for the possibility of agreement occurring by chance alone. Higher Kappa values indicate better agreement beyond chance. McNemar’s Test P-Value: It assesses whether differences in model performance compared to a baseline are statistically significant. A significant p-value (usually < 0.05) suggests that the model performs differently from the baseline.