Video-Game-Metacritic-Scores-Prediction

An attempt to predict Metacritic scores of video games using several machine learning models in Python.
A complete walkthrough of the process can be found in each of the 5 notebooks.

Contents

1. Gathering Data from RWAG API
2. Data Cleaning
3. Video Game Data Analysis
4. Preprocessing Dataset for Prediction
5. Predictive Models Results - Regression
6. Predictive Models Results - Classification

Part 1: Gathering Data From RWAG API

Using the request library, make API calls from the RWAG database to collect video game data. Note: Extract the data.rar file if needed (csv file is too large to push to GitHub)

Part 2: Data Cleaning

Examine the raw data, and trim it down such that the dataset only contains useful information for analysis and prediction.
Data is processed with pandas, numpy and Abstract Syntax Trees (ast).

Part 3: Video Game Data Analysis

Investigate the dataset to derive insights on the most popular games, genres, trends, etc.
Data is visualised using matplotlib, seaborn, plotly and wordcloud.

Part 4: Preprocessing Dataset for Prediction

Prepare the cleaned dataset as an input for machine learning models.
Specifically, the platforms, genres and tags columns (containing lists) are processed into binary columns:

• Platform families are grouped together with numpy
• Tags are cross-checked with the tags scrapped from Steam with fuzzywuzzy and BeautifulSoup

These 3 pieces of information will be used as features in the predictive models.
The train and test sets are then generated with sklearn.
A simple prediction test using the Logistic Regression is made as well.

Part 5: Predictive Models Results - Regression

The following regression models are used, along with their respective RMSE on the test set:

Model	Test RMSE
Linear Regression	169784033252.96356
Random Forest Regressor	9.487118217587387
Support Vector Machine	9.857068901488233
Decision Tree Regression	13.64271924908898
K-Nearest Neighbors	10.592563672173583
AdaBoost	10.62350517308404
Gradient Boost Regressor	10.086508353719507
Extra Tree Regressor	13.182428792510434
XGBoost Regressor	28.081992487362925
Ridge Regression	9.877904001325112

Optimisation is been attempted to fine-tune the parameters of the models. Furthermore, the pycaret library is also used as well:

Model	Test RMSE (with GridSearch)	Test RMSE (Base)	Absolute Error
Random Forest Regressor	9.487118217587387	9.547566374653456	7.317760953625915
Support Vector Machine	9.857068901488233	9.516711070354274	7.132888084897704
PyCaret	-	9.815711126761942	7.570090718963135

Part 6: Predictive Models Results - Classification

The problem could also be tackled as a multi-class classification problem.
A 25-50-75 quartile split was used to establish the cut-off scores for each tier.
The dataset was almost evenly split into these 4 classes:

• Essential - At least 83
• Great - Between 77 & 82
• Good - Between 70 & 76
• Mediocre - At most 69

Both One-Vs-Rest (OvR) and One-Vs-One (OvO) approaches were considered.
The following classification models are used:

Model	F1 Score (OvR)	F1 Score (OvO)
Dummy Classifier	0.26	0.24
Logistic Regression	0.39	0.40
Decision Tree Classifier	0.32	0.35
Random Forest Classifier	0.40	0.40
Extra Tree Classifier	0.40	0.40
Support Vector Machine	0.40	0.40
K-Nearest Neighbors	0.37	0.36
XGBoost Classifier	0.40	0.39

Optimisation is been attempted to fine-tune the parameters of the Random Forest Classifier model.

Model	F1 Score (OvR)	F1 Score (OvO)
Random Forest Classifier	0.40	0.40
Extra Tree Classifier	0.40	0.40
Support Vector Machine	0.40	0.40
Random Forest Classifier (optimised)	0.41	-