Games ESRB Ratings

EDA and prediction exercise

Data: Video Games Rating by ‘ESRB’

Task: Predict the rating for each game in test data

Research Question: What are the the key features that can predict the [ESRB rating] (https://www.esrb.org/ratings-guide/) of a game?

Load libraries

library(tidyverse)
library(skimr)
library(ggsci)
library(Hmisc)
library(corrplot)
library(FactoMineR)
library(factoextra)
library(rpart)
library(rpart.plot)
library(rattle)
library(caret)
library(randomForest)
library(xgboost)
library(pROC)

Import data

train_data= readr::read_csv("Video_games_esrb_rating.csv")

── Column specification ──────────────────────────────────────────────────────────────────────────────────────────────
cols(
  .default = col_double(),
  title = col_character(),
  esrb_rating = col_character()
)
ℹ Use `spec()` for the full column specifications.

test_data= readr::read_csv("test_esrb.csv")

── Column specification ──────────────────────────────────────────────────────────────────────────────────────────────
cols(
  .default = col_double(),
  title = col_character(),
  esrb_rating = col_character()
)
ℹ Use `spec()` for the full column specifications.

Summary

── Data Summary ────────────────────────
                           Values    
Name                       Piped data
Number of rows             1895      
Number of columns          34        
_______________________              
Column type frequency:               
  factor                   34        
________________________             
Group variables            None      

── Variable type: factor ─────────────────────────────────────────────────────────────────────────────────────────────
   skim_variable            n_missing complete_rate ordered n_unique top_counts                     
 1 title                            0             1 FALSE       1890 Bri: 2, Kil: 2, Mut: 2, Mys: 2 
 2 console                          0             1 FALSE          2 0: 994, 1: 901                 
 3 alcohol_reference                0             1 FALSE          2 0: 1800, 1: 95                 
 4 animated_blood                   0             1 FALSE          2 0: 1876, 1: 19                 
 5 blood                            0             1 FALSE          2 0: 1463, 1: 432                
 6 blood_and_gore                   0             1 FALSE          2 0: 1656, 1: 239                
 7 cartoon_violence                 0             1 FALSE          2 0: 1858, 1: 37                 
 8 crude_humor                      0             1 FALSE          2 0: 1792, 1: 103                
 9 drug_reference                   0             1 FALSE          2 0: 1829, 1: 66                 
10 fantasy_violence                 0             1 FALSE          2 0: 1477, 1: 418                
11 intense_violence                 0             1 FALSE          2 0: 1671, 1: 224                
12 language                         0             1 FALSE          2 0: 1690, 1: 205                
13 lyrics                           0             1 FALSE          2 0: 1832, 1: 63                 
14 mature_humor                     0             1 FALSE          2 0: 1873, 1: 22                 
15 mild_blood                       0             1 FALSE          2 0: 1763, 1: 132                
16 mild_cartoon_violence            0             1 FALSE          2 0: 1850, 1: 45                 
17 mild_fantasy_violence            0             1 FALSE          2 0: 1804, 1: 91                 
18 mild_language                    0             1 FALSE          2 0: 1854, 1: 41                 
19 mild_lyrics                      0             1 FALSE          2 0: 1749, 1: 146                
20 mild_suggestive_themes           0             1 FALSE          2 0: 1811, 1: 84                 
21 mild_violence                    0             1 FALSE          2 0: 1805, 1: 90                 
22 no_descriptors                   0             1 FALSE          2 0: 1573, 1: 322                
23 nudity                           0             1 FALSE          2 0: 1867, 1: 28                 
24 partial_nudity                   0             1 FALSE          2 0: 1870, 1: 25                 
25 sexual_content                   0             1 FALSE          2 0: 1830, 1: 65                 
26 sexual_themes                    0             1 FALSE          2 0: 1786, 1: 109                
27 simulated_gambling               0             1 FALSE          2 0: 1768, 1: 127                
28 strong_janguage                  0             1 FALSE          2 0: 1671, 1: 224                
29 strong_sexual_content            0             1 FALSE          2 0: 1827, 1: 68                 
30 suggestive_themes                0             1 FALSE          2 0: 1672, 1: 223                
31 use_of_alcohol                   0             1 FALSE          2 0: 1865, 1: 30                 
32 use_of_drugs_and_alcohol         0             1 FALSE          2 0: 1865, 1: 30                 
33 violence                         0             1 FALSE          2 0: 1774, 1: 121                
34 esrb_rating                      0             1 FALSE          4 T: 689, E: 416, ET: 403, M: 387
── Data Summary ────────────────────────
                           Values    
Name                       Piped data
Number of rows             500       
Number of columns          34        
_______________________              
Column type frequency:               
  factor                   34        
________________________             
Group variables            None      

── Variable type: factor ─────────────────────────────────────────────────────────────────────────────────────────────
   skim_variable            n_missing complete_rate ordered n_unique top_counts                    
 1 title                            0             1 FALSE        494 88 : 2, Cit: 2, Eli: 2, The: 2
 2 console                          0             1 FALSE          2 0: 253, 1: 247                
 3 alcohol_reference                0             1 FALSE          2 0: 474, 1: 26                 
 4 animated_blood                   0             1 FALSE          2 0: 499, 1: 1                  
 5 blood                            0             1 FALSE          2 0: 378, 1: 122                
 6 blood_and_gore                   0             1 FALSE          2 0: 434, 1: 66                 
 7 cartoon_violence                 0             1 FALSE          2 0: 489, 1: 11                 
 8 crude_humor                      0             1 FALSE          2 0: 480, 1: 20                 
 9 drug_reference                   0             1 FALSE          2 0: 484, 1: 16                 
10 fantasy_violence                 0             1 FALSE          2 0: 367, 1: 133                
11 intense_violence                 0             1 FALSE          2 0: 462, 1: 38                 
12 language                         0             1 FALSE          2 0: 435, 1: 65                 
13 lyrics                           0             1 FALSE          2 0: 480, 1: 20                 
14 mature_humor                     0             1 FALSE          1 0: 500                        
15 mild_blood                       0             1 FALSE          2 0: 460, 1: 40                 
16 mild_cartoon_violence            0             1 FALSE          2 0: 494, 1: 6                  
17 mild_fantasy_violence            0             1 FALSE          2 0: 471, 1: 29                 
18 mild_language                    0             1 FALSE          2 0: 473, 1: 27                 
19 mild_lyrics                      0             1 FALSE          2 0: 468, 1: 32                 
20 mild_suggestive_themes           0             1 FALSE          2 0: 480, 1: 20                 
21 mild_violence                    0             1 FALSE          2 0: 486, 1: 14                 
22 no_descriptors                   0             1 FALSE          2 0: 443, 1: 57                 
23 nudity                           0             1 FALSE          2 0: 489, 1: 11                 
24 partial_nudity                   0             1 FALSE          2 0: 471, 1: 29                 
25 sexual_content                   0             1 FALSE          2 0: 490, 1: 10                 
26 sexual_themes                    0             1 FALSE          2 0: 495, 1: 5                  
27 simulated_gambling               0             1 FALSE          2 0: 489, 1: 11                 
28 strong_janguage                  0             1 FALSE          2 0: 474, 1: 26                 
29 strong_sexual_content            0             1 FALSE          2 0: 475, 1: 25                 
30 suggestive_themes                0             1 FALSE          2 0: 462, 1: 38                 
31 use_of_alcohol                   0             1 FALSE          2 0: 457, 1: 43                 
32 use_of_drugs_and_alcohol         0             1 FALSE          2 0: 475, 1: 25                 
33 violence                         0             1 FALSE          2 0: 346, 1: 154                
34 esrb_rating                      0             1 FALSE          4 T: 184, ET: 126, E: 100, M: 90

Target variable

Hmisc::describe(as.factor(train_data$esrb_rating))

as.factor(train_data$esrb_rating) 
       n  missing distinct 
    1895        0        4 
                                  
Value          E    ET     M     T
Frequency    416   403   387   689
Proportion 0.220 0.213 0.204 0.364

Hmisc::describe(as.factor(test_data$esrb_rating))

as.factor(test_data$esrb_rating) 
       n  missing distinct 
     500        0        4 
                                  
Value          E    ET     M     T
Frequency    100   126    90   184
Proportion 0.200 0.252 0.180 0.368

• both test and train data contains four types of ESRB ratings i.e., Everyone (E), Everyone 10+ (ET), Teen (T) and Mature (M), with similar proportions.

prop_train = train_data %>% 
  group_by(esrb_rating) %>% 
  tally() %>%
  mutate(train=round(n/sum(n),3))

prop_test = test_data %>% 
  group_by(esrb_rating) %>% 
  tally() %>%
  mutate(test=round(n/sum(n),3))

rat_p = merge(prop_train, prop_test, by="esrb_rating") %>% #merge 
  mutate(esrb_rating_long = recode(esrb_rating, #rating name
                         "T"="Teen",
                         "E" ="Everyone",
                         "ET" ="Everyone 10+",
                         "M" ="Mature"))  %>%
  select(esrb_rating_long, train, test) %>%
  pivot_longer(!esrb_rating_long, names_to="data", values_to="pct") %>% #wide to long
  # plot
  ggplot(aes(x=esrb_rating_long, y=pct, fill= data)) + 
  geom_col(position="dodge") + 
  geom_text(aes(label=paste0(pct*100,"%")), position=position_dodge(width=1), vjust=1.5, color="white", size=3.5) +
  scale_x_discrete(limits=c("Everyone","Everyone 10+","Teen","Mature")) +
  scale_fill_manual(values=c("#30638e","#003d5b")) +
  scale_y_continuous(limits=c(0,0.37), labels=scales::percent) + 
  theme_bw() + 
  theme(plot.title=element_text(size=11)) +
  labs(fill="Data",
       y="Percentage",
       x="ESRB Rating",
       title="Percent distribution of ESRB rating in test and train data")
rat_p 

# recode esrb rating to numeric for classification
train_data %>% 
  mutate(esrb = recode(esrb_rating, 
                "E" = 0,
                "ET"= 1,
                "T" =2,
                "M" =3)) -> train_data 

test_data %>% 
  mutate(esrb = recode(esrb_rating, 
                "E" = 0,
                "ET"= 1,
                "T" =2,
                "M" =3)) -> test_data

Correlation

# function to format matrix
flattenCorrMatrix <- function(cormat, pmat) {
  ut <- upper.tri(cormat)
  data.frame(
    row = rownames(cormat)[row(cormat)[ut]],
    column = rownames(cormat)[col(cormat)[ut]],
    cor  =(cormat)[ut],
    p = pmat[ut]
    )
}

# train data 
train_features = train_data %>% select(where(is.numeric))
res_train = rcorr(as.matrix(train_features), type=c("spearman"))
cor_train = flattenCorrMatrix(res_train$r, res_train$P)

# sig. corr (highest 10)
cor_train %>% 
  mutate(across(3:4, round,4)) %>%
  filter(column=="esrb") %>% 
  filter(p<0.01) %>% 
  arrange(desc(cor)) %>% 
  dplyr::slice(1:10)

# test data 
test_features = test_data %>% select(where(is.numeric))
res_test = rcorr(as.matrix(test_features), type=c("spearman"))
cor_test = flattenCorrMatrix(res_test$r, res_test$P)

# sig. corr (highest 10)
cor_test %>% 
  mutate(across(3:4, round,4)) %>%
  filter(column=="esrb") %>% 
  filter(p<0.05) %>% 
  arrange(desc(cor)) %>% 
  dplyr::slice(1:10)

• blood_and_gore and blood have a correlation of more than 0.4 to esrb_rating, in both test and train data

# correlation heatmap
# reference: # http://www.sthda.com/english/wiki/visualize-correlation-matrix-using-correlogram

# function for pvalue
cor.mtest <- function(mat, ...) {
    mat <- as.matrix(mat)
    n <- ncol(mat)
    p.mat<- matrix(NA, n, n)
    diag(p.mat) <- 0
    for (i in 1:(n - 1)) {
        for (j in (i + 1):n) {
            tmp <- cor.test(mat[, i], mat[, j], ...)
            p.mat[i, j] <- p.mat[j, i] <- tmp$p.value
        }
    }
  colnames(p.mat) <- rownames(p.mat) <- colnames(mat)
  p.mat
}

# plot train data
c_train = cor(train_features, method=c("spearman"))
p.mat <- cor.mtest(train_features) # pvalue

corrplot(c_train, method="color",order="hclust", 
         p.mat = p.mat, sig.level = 0.01, insig = "blank", 
         tl.col="black", tl.cex=0.7,
         diag=FALSE 
         )

Visualise ratings by group (PCA)

• Reference: http://www.sthda.com/english/articles/31-principal-component-methods-in-r-practical-guide/112-pca-principal-component-analysis-essentials/

train_label = train_data %>% mutate(label= paste(esrb,":",esrb_rating))

pca_train <- PCA(train_features[,-33], graph = FALSE)

fviz_pca_ind(pca_train,
             geom.ind = "point", # show points only
             col.ind = as.factor(train_label$label), # color by rating
             alpha.ind=0.8,
             palette = "jco",
             #addEllipses = TRUE, # Concentration ellipses
             legend.title = "ESRB Rating",
             title= "Train data"
             )

test_label = test_data %>% mutate(label= paste(esrb,":",esrb_rating))

pca_test <- PCA(test_features[,-33], graph = FALSE)

fviz_pca_ind(pca_test,
             geom.ind = "point", # show points only
             col.ind = as.factor(test_label$label), # color by rating
             alpha.ind=0.8,
             palette = "jco",
             #addEllipses = TRUE, # Concentration ellipses
             legend.title = "ESRB Rating",
             title= "Test data"
             )

Prediction

• Using all the variables in the dataset to find the key features that can predict the game’s ERSB rating.

Decision Tree

dt = rpart(esrb~., data= train_features, method="class")
fancyRpartPlot(dt)

printcp(dt)

Classification tree:
rpart(formula = esrb ~ ., data = train_features, method = "class")

Variables actually used in tree construction:
[1] blood                 blood_and_gore        fantasy_violence      mild_fantasy_violence no_descriptors       
[6] strong_janguage       suggestive_themes    

Root node error: 1206/1895 = 0.63641

n= 1895 

        CP nsplit rel error  xerror     xstd
1 0.183250      0   1.00000 1.00000 0.017363
2 0.111940      2   0.63350 0.63350 0.017706
3 0.063018      3   0.52156 0.52156 0.016998
4 0.048922      4   0.45854 0.45854 0.016409
5 0.017413      5   0.40962 0.40962 0.015846
6 0.012438      6   0.39221 0.39221 0.015622
7 0.010000      7   0.37977 0.38640 0.015544

plotcp(dt)

dt$variable.importance

      strong_janguage        no_descriptors      fantasy_violence mild_fantasy_violence        blood_and_gore 
          203.0573083           188.2246279           101.7050377            79.2844795            70.7085658 
                blood     suggestive_themes strong_sexual_content            mild_blood                lyrics 
           23.3380036            18.1659637            13.9615640             9.0522808             1.0845351 
       sexual_content     alcohol_reference 
            0.5422676             0.2711338 

# confusion matrix
dt.p = predict(dt, test_features, type = "class")
confusionMatrix(dt.p, as.factor(test_features$esrb))

Confusion Matrix and Statistics

          Reference
Prediction   0   1   2   3
         0  74   4   3   0
         1   0  93  27   1
         2  26  29 142  25
         3   0   0  12  64

Overall Statistics
                                          
               Accuracy : 0.746           
                 95% CI : (0.7055, 0.7836)
    No Information Rate : 0.368           
    P-Value [Acc > NIR] : < 2.2e-16       
                                          
                  Kappa : 0.6452          
                                          
 Mcnemar's Test P-Value : NA              

Statistics by Class:

                     Class: 0 Class: 1 Class: 2 Class: 3
Sensitivity            0.7400   0.7381   0.7717   0.7111
Specificity            0.9825   0.9251   0.7468   0.9707
Pos Pred Value         0.9136   0.7686   0.6396   0.8421
Neg Pred Value         0.9379   0.9129   0.8489   0.9387
Prevalence             0.2000   0.2520   0.3680   0.1800
Detection Rate         0.1480   0.1860   0.2840   0.1280
Detection Prevalence   0.1620   0.2420   0.4440   0.1520
Balanced Accuracy      0.8613   0.8316   0.7593   0.8409

# AUC
multiclass.roc(response= test_features$esrb, predictor = factor(dt.p, ordered=TRUE), plot=FALSE, print.auc=TRUE)

Call:
multiclass.roc.default(response = test_features$esrb, predictor = factor(dt.p,     ordered = TRUE), plot = FALSE, print.auc = TRUE)

Data: factor(dt.p, ordered = TRUE) with 4 levels of test_features$esrb: 0, 1, 2, 3.
Multi-class area under the curve: 0.8634

# pruning
dt_prune=prune(dt, cp=0.029)
fancyRpartPlot(dt_prune)

printcp(dt_prune)

Classification tree:
rpart(formula = esrb ~ ., data = train_features, method = "class")

Variables actually used in tree construction:
[1] blood_and_gore        fantasy_violence      mild_fantasy_violence no_descriptors        strong_janguage      

Root node error: 1206/1895 = 0.63641

n= 1895 

        CP nsplit rel error  xerror     xstd
1 0.183250      0   1.00000 1.00000 0.017363
2 0.111940      2   0.63350 0.63350 0.017706
3 0.063018      3   0.52156 0.52156 0.016998
4 0.048922      4   0.45854 0.45854 0.016409
5 0.029000      5   0.40962 0.40962 0.015846

dt_prune$variable.importance

      strong_janguage        no_descriptors      fantasy_violence mild_fantasy_violence        blood_and_gore 
           203.057308            188.224628            101.705038             79.284479             70.708566 
strong_sexual_content            mild_blood 
            13.961564              9.052281 

# confusion matrix 
dt_prune.p = predict(dt_prune, test_features, type = "class")
confusionMatrix(dt_prune.p, as.factor(test_features$esrb))

Confusion Matrix and Statistics

          Reference
Prediction   0   1   2   3
         0  74   4   6   0
         1   0  94  27   2
         2  26  28 139  24
         3   0   0  12  64

Overall Statistics
                                          
               Accuracy : 0.742           
                 95% CI : (0.7013, 0.7798)
    No Information Rate : 0.368           
    P-Value [Acc > NIR] : < 2.2e-16       
                                          
                  Kappa : 0.6403          
                                          
 Mcnemar's Test P-Value : NA              

Statistics by Class:

                     Class: 0 Class: 1 Class: 2 Class: 3
Sensitivity            0.7400   0.7460   0.7554   0.7111
Specificity            0.9750   0.9225   0.7532   0.9707
Pos Pred Value         0.8810   0.7642   0.6406   0.8421
Neg Pred Value         0.9375   0.9151   0.8410   0.9387
Prevalence             0.2000   0.2520   0.3680   0.1800
Detection Rate         0.1480   0.1880   0.2780   0.1280
Detection Prevalence   0.1680   0.2460   0.4340   0.1520
Balanced Accuracy      0.8575   0.8342   0.7543   0.8409

# AUC
multiclass.roc(response= test_features$esrb, predictor = factor(dt_prune.p, ordered=TRUE), plot=FALSE, print.auc=TRUE)

Call:
multiclass.roc.default(response = test_features$esrb, predictor = factor(dt_prune.p,     ordered = TRUE), plot = FALSE, print.auc = TRUE)

Data: factor(dt_prune.p, ordered = TRUE) with 4 levels of test_features$esrb: 0, 1, 2, 3.
Multi-class area under the curve: 0.8587

# plot dt variable importance
dtdf1 <- data.frame(imp = dt$variable.importance)
dtdf2 <- dtdf1 %>% 
  tibble::rownames_to_column() %>% 
  dplyr::rename("variable" = rowname) %>% 
  dplyr::arrange(imp) %>%
  dplyr::mutate(variable = forcats::fct_inorder(variable))
ggplot2::ggplot(dtdf2) +
  geom_point(aes(x = variable, y = imp), show.legend = F, shape=1, size=3) +
  coord_flip() +
  theme_bw() + 
  theme(axis.ticks.y = element_blank(),
        panel.grid.major.x=element_blank(),
        panel.grid.minor.x=element_blank(),
        panel.grid=element_line(linetype = "dotted",size=0.7, color="grey"),
        plot.title=element_text(hjust=0.5, face="bold"),
        plot.margin=ggplot2::margin(10,40,10,30)) +
  labs(title="dt")

Random forest

set.seed(123)
rf2 <- train(factor(esrb) ~., data = train_features, method = "rf",
  trControl = trainControl("cv", number = 10),
  importance = TRUE)

rf2$bestTune

rf2$finalModel

Call:
 randomForest(x = x, y = y, mtry = param$mtry, importance = TRUE) 
               Type of random forest: classification
                     Number of trees: 500
No. of variables tried at each split: 32

        OOB estimate of  error rate: 14.67%
Confusion matrix:
    0   1   2   3 class.error
0 404  10   2   0  0.02884615
1  19 328  55   1  0.18610422
2   7  91 550  41  0.20174165
3   1   0  51 335  0.13436693

# predict
rf2.p <- rf2 %>% predict(test_features)
# accuracy
mean(rf2.p == test_features$esrb)

[1] 0.846

# confusion matrix
confusionMatrix(rf2.p, as.factor(test_features$esrb))

Confusion Matrix and Statistics

          Reference
Prediction   0   1   2   3
         0  95   4   2   0
         1   4 112  19   2
         2   1  10 155  27
         3   0   0   8  61

Overall Statistics
                                          
               Accuracy : 0.846           
                 95% CI : (0.8113, 0.8765)
    No Information Rate : 0.368           
    P-Value [Acc > NIR] : < 2.2e-16       
                                          
                  Kappa : 0.7872          
                                          
 Mcnemar's Test P-Value : NA              

Statistics by Class:

                     Class: 0 Class: 1 Class: 2 Class: 3
Sensitivity            0.9500   0.8889   0.8424   0.6778
Specificity            0.9850   0.9332   0.8797   0.9805
Pos Pred Value         0.9406   0.8175   0.8031   0.8841
Neg Pred Value         0.9875   0.9614   0.9055   0.9327
Prevalence             0.2000   0.2520   0.3680   0.1800
Detection Rate         0.1900   0.2240   0.3100   0.1220
Detection Prevalence   0.2020   0.2740   0.3860   0.1380
Balanced Accuracy      0.9675   0.9110   0.8611   0.8291

# AUC
multiclass.roc(response= test_features$esrb, predictor = factor(rf2.p, ordered=TRUE), plot=FALSE, print.auc=TRUE)

Call:
multiclass.roc.default(response = test_features$esrb, predictor = factor(rf2.p,     ordered = TRUE), plot = FALSE, print.auc = TRUE)

Data: factor(rf2.p, ordered = TRUE) with 4 levels of test_features$esrb: 0, 1, 2, 3.
Multi-class area under the curve: 0.9407

# variable importance
varImpPlot(rf2$finalModel, type=2)

#varImp(rf2) 
#importance(rf2$finalModel)

Gradient boosting

set.seed(123)
xgb <- train(factor(esrb) ~., data = train_features, method = "xgbTree",trControl = trainControl("cv", number = 10))
xgb$bestTune

xgb.p = xgb %>% predict(test_features)
mean(xgb.p == test_features$esrb)

[1] 0.828

# confusion matrix
confusionMatrix(xgb.p, as.factor(test_features$esrb))

Confusion Matrix and Statistics

          Reference
Prediction   0   1   2   3
         0  95   4   1   0
         1   4 103  18   0
         2   1  19 159  33
         3   0   0   6  57

Overall Statistics
                                         
               Accuracy : 0.828          
                 95% CI : (0.792, 0.8601)
    No Information Rate : 0.368          
    P-Value [Acc > NIR] : < 2.2e-16      
                                         
                  Kappa : 0.7605         
                                         
 Mcnemar's Test P-Value : NA             

Statistics by Class:

                     Class: 0 Class: 1 Class: 2 Class: 3
Sensitivity            0.9500   0.8175   0.8641   0.6333
Specificity            0.9875   0.9412   0.8323   0.9854
Pos Pred Value         0.9500   0.8240   0.7500   0.9048
Neg Pred Value         0.9875   0.9387   0.9132   0.9245
Prevalence             0.2000   0.2520   0.3680   0.1800
Detection Rate         0.1900   0.2060   0.3180   0.1140
Detection Prevalence   0.2000   0.2500   0.4240   0.1260
Balanced Accuracy      0.9688   0.8793   0.8482   0.8093

# variable importance (pct)
varImp(xgb)

xgbTree variable importance

  only 20 most important variables shown (out of 32)

Summary

• Most important features:
  • Decision tree: strong_janguage, no_descriptors, fantasy_violence, mild_fantasy_violence, blood and gore
    
  • Random Forest: strong_janguage, no_descriptors, fantasy_violence, blood, blood_and_gore,
    
  • XGBoost: strong_janguage, blood, no_descriptors, fantasy_violance, blood_and gore
• What are the the key features that can predict the ESRB rating of a game?
  • Decision tree, random forest and XGBoost models shows that strong language, blood and gore, no descriptors and fantasy violance are key features in predicting the games’ ESRB rating.