GAME RATINGS
PREDICTOR
Video games ESRB (Entertainment Software
Rating Board) predictor
Control Systems and Computer Engineering – Sapienza University of Rome
Quantitative Methods for Computer Science
By Roberto Falconi and Federico Guidi
INTRODUCTION
►
►
►
Roberto Falconi
Federico Guidi
PROCEDURE
►
►
►
►
►
►
►
►
Roberto Falconi
Federico Guidi
Roberto Falconi
Federico Guidi
Super Mario Sunshine
2002 (E)
Grand Theft Auto V
2013 (M)
The Legend of Zelda
Breath of the Wild
2017 (E10+)
Uncharted 4
2016 (T)
Roberto Falconi
Federico Guidi
DATASET ANALYSIS
DATASET SLICE
Roberto Falconi
Federico Guidi
DATASET CLASSES RIPARTITION
Number of elements per class
Everyone Mature Everyone 10+ Teen Roberto Falconi
Federico Guidi
DATASET ANALYSIS
SALES RATING’S IMPORTANCE
Roberto Falconi
Federico Guidi
DATASET ANALYSIS
SALES RATING’S IMPORTANCE
Roberto Falconi
Federico Guidi
DATASET ANALYSIS
SALES RATING’S IMPORTANCE
Roberto Falconi
Federico Guidi
SETUP
UBUNTU, DEBIAN E MACOS
Roberto Falconi
Federico Guidi
DATASET CONFIGURATION
INCOMPLETE ELEMENTS DELETION
Name Rating
Super
Mario
E
FIFA T
Pokémon E10
Tetris NaN
Name Rating
Super
Mario
E
FIFA T
Pokémon E10
Roberto Falconi
Federico Guidi
DATASET CONFIGURATION
APPLYING ONE-HOT ENCODING
Name Rating
Super
Mario
E
FIFA T
Pokémon E10
Name Rating_E Rating_E10 Rating_T
Super
Mario
1 0 0
FIFA 0 0 1
Pokémon 0 1 0
Roberto Falconi
Federico Guidi
DATASET CONFIGURATION
TRAINING SET AND TEST SET
Name Rating_E Rating_E10 Rating_T
Pokémon 0 1 0
Name Rating_E Rating_E10 Rating_T
Super
Mario
1 0 0
FIFA 0 0 1
Name Rating_E Rating_E10 Rating_T
Super
Mario
1 0 0
FIFA 0 0 1
Pokémon 0 1 0
Roberto Falconi
Federico Guidi
DATASET CONFIGURATION
TRAINING SET AND TEST SET
▶
▶
▶
Roberto Falconi
Federico Guidi
LOGISTIC REGRESSION
Roberto Falconi
Federico Guidi
LOGISTIC REGRESSION
▶
Roberto Falconi
Federico Guidi
LOGISTIC REGRESSION
PYTHON CODE
Roberto Falconi
Federico Guidi
LOGISTIC REGRESSION
▶
▶
▶
▶
Pros
▶
▶
▶
▶
Cons
Roberto Falconi
Federico Guidi
RANDOM FOREST
Roberto Falconi
Federico Guidi
RANDOM FOREST
▶
B Roberto Falconi
Federico Guidi
RANDOM FOREST
PYTHON CODE
Roberto Falconi
Federico Guidi
RANDOM FOREST
▶
▶
▶
▶
▶
▶
▶
Pro
▶
Contro
Roberto Falconi
Federico Guidi
K-NN
Roberto Falconi
Federico Guidi
K-NN
▶
Roberto Falconi
Federico Guidi
K-NN
PYTHON CODE
Roberto Falconi
Federico Guidi
K-NN
▶
▶
▶
Pro
Contro
▶
▶
▶
▶
▶ Roberto Falconi
Federico Guidi
RUNNING CLASSIFICATORS
70%
75%
80%
85%
90%
E E10 T M
Accuracy Score / Cross-validation
Logistic Regression Random Forest k-NN
Roberto Falconi
Federico Guidi
▶
▶
▶
RUNNING CLASSIFICATORS
Roberto Falconi
Federico Guidi
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Elemento 1 Elemento 2 Elemento 3 Elemento 4
Random Forest - confidence
(probability that an element belongs to a class)
E E10 T M
RUNNING CLASSIFICATORS
Roberto Falconi
Federico Guidi
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Elemento 1 Elemento 2 Elemento 3 Elemento 4
Random Forest – normalized confidence
(probability that an element belongs to a class )
E E10 T M
RUNNING CLASSIFICATORS
Roberto Falconi
Federico Guidi
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
Elemento 1 Elemento 2 Elemento 3 Elemento 4
Logistic Regression - confidence
(probability that an element belongs to a class )
E E10 T M
RUNNING CLASSIFICATORS
Roberto Falconi
Federico Guidi
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Elemento 1 Elemento 2 Elemento 3 Elemento 4
Logistic Regression – normalized confidence
(probability that an element belongs to a class )
E E10 T M
Misclassification on element 1
RUNNING CLASSIFICATORS
Roberto Falconi
Federico Guidi
0%
10%
20%
30%
40%
50%
60%
70%
80%
Elemento 1 Elemento 2 Elemento 3 Elemento 4
k-NN - confidence
(probability that an element belongs to a class )
E E10 T M
RUNNING CLASSIFICATORS
Roberto Falconi
Federico Guidi
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Elemento 1 Elemento 2 Elemento 3 Elemento 4
k-NN – normalized confidence
(probability that an element belongs to a class )
E E10 T M
RUNNING CLASSIFICATORS
Roberto Falconi
Federico GuidiMisclassification on element 2 and element 3
BIAS-VARIANCE TRADEOFF
OBSERVATIONS
Roberto Falconi
Federico Guidi
BIAS-VARIANCE TRADEOFF
OBSERVATIONS
▶
▶
▶
Roberto Falconi
Federico Guidi
CONCLUSION
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Random Forest Logistic Regression k-NN
Final scores
Roberto Falconi
Federico Guidi
CONCLUSION
Name Rating
Madden NFL E
Mafia III M
No Man’s Sky T
NBA 2K17 E
Slice of dataset Output
Roberto Falconi
Federico Guidi

Game Ratings Predictor - machine learning software to predict video games content rating