VideoGames
This project is the continuation of project part 1. Link for project part 1: https://rpubs.com/S_ubin10/809024
All library and packages required for the project are listed below:
library(readxl)
library(rvest)
library(tidyverse)## ── Attaching packages ─────────────────────────────────────── tidyverse 1.3.1 ──## ✓ ggplot2 3.3.5 ✓ purrr 0.3.4
## ✓ tibble 3.1.4 ✓ dplyr 1.0.7
## ✓ tidyr 1.1.3 ✓ stringr 1.4.0
## ✓ readr 2.0.1 ✓ forcats 0.5.1## ── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──
## x dplyr::filter() masks stats::filter()
## x readr::guess_encoding() masks rvest::guess_encoding()
## x dplyr::lag() masks stats::lag()library(dplyr)
install.packages("rpart.plot")## Installing package into '/cloud/lib/x86_64-pc-linux-gnu-library/4.1'
## (as 'lib' is unspecified)df = read_excel("videogames.xlsx") Loading the list of data set contains a list of video games with sales greater than 100,000 copies. Each row in the data set represents a description of a video game.
head(df)## # A tibble: 6 × 11
## Rank Name Platform Year Genre Publisher NA_Sales EU_Sales JP_Sales
## <dbl> <chr> <chr> <chr> <chr> <chr> <dbl> <dbl> <dbl>
## 1 1 Wii Sports Wii 2006 Sports Nintendo 41.5 29.0 3.77
## 2 2 Super Mario… NES 1985 Platfo… Nintendo 29.1 3.58 6.81
## 3 3 Mario Kart … Wii 2008 Racing Nintendo 15.8 12.9 3.79
## 4 4 Wii Sports … Wii 2009 Sports Nintendo 15.8 11.0 3.28
## 5 5 Pokemon Red… GB 1996 Role-P… Nintendo 11.3 8.89 10.2
## 6 6 Tetris GB 1989 Puzzle Nintendo 23.2 2.26 4.22
## # … with 2 more variables: Other_Sales <dbl>, Global_Sales <dbl>Here we a sample of 150 observations from 16598 observations. For decision tree, we take one categorical variable, Platform in this case and two quantitative variables (JP_Sale and NA_Sales) as a factors to make a decision tree.
smalldf <- sample_n (df,150)
library(rpart)
library(rpart.plot)
tree <- rpart(Platform ~ JP_Sales + NA_Sales,data = smalldf)
tree## n= 150
##
## node), split, n, loss, yval, (yprob)
## * denotes terminal node
##
## 1) root 150 126 PS2 (0.02 0.033 0.13 0.0067 0.06 0.027 0.0067 0.053 0.093 0.16 0.087 0.033 0.067 0.013 0.0067 0.0067 0.08 0.06 0.047 0.0067)
## 2) NA_Sales< 0.115 84 68 DS (0 0.06 0.19 0 0.024 0.024 0 0.095 0.11 0.11 0.048 0.024 0.12 0.024 0.012 0 0.071 0.024 0.071 0)
## 4) JP_Sales>=0.115 8 3 PS (0 0 0.12 0 0 0 0 0 0.62 0.13 0 0 0.12 0 0 0 0 0 0 0) *
## 5) JP_Sales< 0.115 76 61 DS (0 0.066 0.2 0 0.026 0.026 0 0.11 0.053 0.11 0.053 0.026 0.12 0.026 0.013 0 0.079 0.026 0.079 0)
## 10) JP_Sales>=0.005 28 21 DS (0 0.18 0.25 0 0 0 0 0 0 0.071 0.036 0.036 0.25 0.071 0.036 0 0 0.036 0.036 0)
## 20) JP_Sales>=0.035 13 8 DS (0 0.23 0.38 0 0 0 0 0 0 0.077 0 0 0.23 0 0.077 0 0 0 0 0) *
## 21) JP_Sales< 0.035 15 11 PSP (0 0.13 0.13 0 0 0 0 0 0 0.067 0.067 0.067 0.27 0.13 0 0 0 0.067 0.067 0) *
## 11) JP_Sales< 0.005 48 40 DS (0 0 0.17 0 0.042 0.042 0 0.17 0.083 0.12 0.062 0.021 0.042 0 0 0 0.12 0.021 0.1 0)
## 22) NA_Sales< 0.005 9 3 PC (0 0 0.22 0 0 0 0 0.67 0 0 0.11 0 0 0 0 0 0 0 0 0) *
## 23) NA_Sales>=0.005 39 33 DS (0 0 0.15 0 0.051 0.051 0 0.051 0.1 0.15 0.051 0.026 0.051 0 0 0 0.15 0.026 0.13 0)
## 46) NA_Sales>=0.085 11 6 DS (0 0 0.45 0 0 0.091 0 0 0.091 0.18 0 0 0 0 0 0 0.091 0 0.091 0) *
## 47) NA_Sales< 0.085 28 23 Wii (0 0 0.036 0 0.071 0.036 0 0.071 0.11 0.14 0.071 0.036 0.071 0 0 0 0.18 0.036 0.14 0) *
## 3) NA_Sales>=0.115 66 51 PS2 (0.045 0 0.061 0.015 0.11 0.03 0.015 0 0.076 0.23 0.14 0.045 0 0 0 0.015 0.091 0.11 0.015 0.015)
## 6) NA_Sales>=0.255 46 39 PS3 (0.043 0 0.065 0.022 0.13 0.043 0.022 0 0.087 0.11 0.15 0.065 0 0 0 0.022 0.087 0.13 0.022 0)
## 12) NA_Sales< 0.55 30 23 PS3 (0.067 0 0.067 0 0.17 0 0 0 0.13 0.067 0.23 0.033 0 0 0 0 0.1 0.1 0.033 0)
## 24) NA_Sales>=0.325 21 14 PS3 (0.095 0 0.048 0 0.19 0 0 0 0.095 0.048 0.33 0 0 0 0 0 0.14 0 0.048 0) *
## 25) NA_Sales< 0.325 9 6 X360 (0 0 0.11 0 0.11 0 0 0 0.22 0.11 0 0.11 0 0 0 0 0 0.33 0 0) *
## 13) NA_Sales>=0.55 16 13 PS2 (0 0 0.062 0.063 0.062 0.13 0.063 0 0 0.19 0 0.12 0 0 0 0.063 0.063 0.19 0 0) *
## 7) NA_Sales< 0.255 20 10 PS2 (0.05 0 0.05 0 0.05 0 0 0 0.05 0.5 0.1 0 0 0 0 0 0.1 0.05 0 0.05) *Creating a visualization for a decision tree.
rpart.plot(tree, extra = 2)## Warning: All boxes will be white (the box.palette argument will be ignored) because
## the number of classes in the response 20 is greater than length(box.palette) 6.
## To silence this warning use box.palette=0 or trace=-1.
Making a prediction for a following variable to compate the actual value.
pred <- predict (tree,df,type = "class")
head(pred)## 1 2 3 4 5 6
## PS2 PS2 PS2 PS2 PS2 PS2
## 20 Levels: 2600 3DS DS GB GBA GC NES PC PS PS2 PS3 PS4 PSP PSV SAT SNES ... XOneProbabilities of Classification
predict(tree,smalldf)%>%
head()## 2600 3DS DS GB GBA GC NES PC PS PS2
## 1 0.0952381 0.0000000 0.04761905 0 0.1904762 0 0 0 0.0952381 0.04761905
## 2 0.0000000 0.2307692 0.38461538 0 0.0000000 0 0 0 0.0000000 0.07692308
## 3 0.0000000 0.0000000 0.12500000 0 0.0000000 0 0 0 0.6250000 0.12500000
## 4 0.0000000 0.0000000 0.12500000 0 0.0000000 0 0 0 0.6250000 0.12500000
## 5 0.0000000 0.0000000 0.11111111 0 0.1111111 0 0 0 0.2222222 0.11111111
## 6 0.0500000 0.0000000 0.05000000 0 0.0500000 0 0 0 0.0500000 0.50000000
## PS3 PS4 PSP PSV SAT SNES Wii X360
## 1 0.3333333 0.0000000 0.0000000 0 0.00000000 0 0.1428571 0.0000000
## 2 0.0000000 0.0000000 0.2307692 0 0.07692308 0 0.0000000 0.0000000
## 3 0.0000000 0.0000000 0.1250000 0 0.00000000 0 0.0000000 0.0000000
## 4 0.0000000 0.0000000 0.1250000 0 0.00000000 0 0.0000000 0.0000000
## 5 0.0000000 0.1111111 0.0000000 0 0.00000000 0 0.0000000 0.3333333
## 6 0.1000000 0.0000000 0.0000000 0 0.00000000 0 0.1000000 0.0500000
## XB XOne
## 1 0.04761905 0.00
## 2 0.00000000 0.00
## 3 0.00000000 0.00
## 4 0.00000000 0.00
## 5 0.00000000 0.00
## 6 0.00000000 0.05When inspecting a classification model’s performance, a confusion table tells you the distribution of the predictions and targets. Making confusion table..
confusion_table <- with(df,table(Genre,pred))
confusion_table## pred
## Genre 2600 3DS DS GB GBA GC NES PC PS PS2 PS3 PS4 PSP
## Action 0 0 482 0 0 0 0 168 180 1079 295 0 272
## Adventure 0 0 263 0 0 0 0 85 72 156 40 0 401
## Fighting 0 0 159 0 0 0 0 5 116 237 76 0 77
## Misc 0 0 288 0 0 0 0 120 131 505 126 0 154
## Platform 0 0 89 0 0 0 0 24 38 348 99 0 17
## Puzzle 0 0 70 0 0 0 0 58 42 121 30 0 16
## Racing 0 0 134 0 0 0 0 83 28 391 113 0 39
## Role-Playing 0 0 299 0 0 0 0 61 300 353 73 0 171
## Shooter 0 0 110 0 0 0 0 77 40 475 131 0 48
## Simulation 0 0 101 0 0 0 0 87 72 214 82 0 66
## Sports 0 0 292 0 0 0 0 143 217 785 246 0 66
## Strategy 0 0 101 0 0 0 0 130 111 90 20 0 58
## pred
## Genre PSV SAT SNES Wii X360 XB XOne
## Action 0 0 0 669 171 0 0
## Adventure 0 0 0 246 23 0 0
## Fighting 0 0 0 131 47 0 0
## Misc 0 0 0 346 69 0 0
## Platform 0 0 0 227 44 0 0
## Puzzle 0 0 0 220 25 0 0
## Racing 0 0 0 406 55 0 0
## Role-Playing 0 0 0 182 49 0 0
## Shooter 0 0 0 360 69 0 0
## Simulation 0 0 0 200 45 0 0
## Sports 0 0 0 479 118 0 0
## Strategy 0 0 0 161 10 0 0We used a Genre as the variable to make the confusion table.
Cross Validation
s<-sample(140,115)
train_data <- df[s,]
test_data <- df[-s,]Now using training data to build model and then test it.
dim(train_data)## [1] 115 11dim(test_data)## [1] 16483 11Now creating a full tree for following data.
dtm<-rpart(Platform ~ JP_Sales, train_data,method ="class" )
rpart.plot(dtm,extra =2)## Warning: All boxes will be white (the box.palette argument will be ignored) because
## the number of classes in the response 18 is greater than length(box.palette) 6.
## To silence this warning use box.palette=0 or trace=-1.
In this part we create chi-square table to examine the importance of each feature.
library(FSelector)
weights <- df %>% chi.squared (Platform~ .,data = .) %>%
as_tibble(rownames = "feature")%>%
arrange(desc(attr_importance))
weights ## # A tibble: 10 × 2
## feature attr_importance
## <chr> <dbl>
## 1 Name 0.872
## 2 Year 0.387
## 3 Publisher 0.385
## 4 JP_Sales 0.330
## 5 EU_Sales 0.292
## 6 NA_Sales 0.271
## 7 Other_Sales 0.244
## 8 Rank 0.234
## 9 Global_Sales 0.233
## 10 Genre 0.180According to the table, Genre has the importace of 0.1799156 which is the least among all the features and Name has the importance of 0.8722909 which is the highest among all.
ggplot(weights,
aes(x=attr_importance, y =reorder(feature, attr_importance)))+
geom_bar(stat="identity") +
xlab("Importance score")+ ylab("Feature")
To make the decision tree, I only used three variables because compiling the tree using all the data was taking forever but I used all the variables to make the table above. So, comparing the decision tree and chi-square statistics accurately is difficult.