COS 397 - Machine Learning and Data Science

Github Repo for an Independent Work Seminar at Princeton

Predicting Pokémon Type and Generation Using Machine Learning Techniques

This project focuses on predicting Pokemon types and generations using various machine learning models. So far I have implemented and compared several algorithms to classify Pokemon based on their stats and characteristics.

Project Overview

Data: I am using a dataset of Pokemon with features like HP, Attack, Defense, Special Attack, Special Defense, Speed, Height, and Weight. The dataset is scraped from https://pokemondb.net/pokedex/all.

Prediction Tasks:

• Primary Type Prediction
• Both Types Prediction (Multi-label classification)
• Generation Prediction

Models Implemented

I have experimented with several machine learning models so far:

1. Logistic Regression
2. K-Nearest Neighbors (KNN)
3. Random Forests
4. Support Vector Machines (SVM) - both linear and RBF kernels
5. Gradient Boosting (XGBoost)

Methodology

For each model, the general implementation steps are as follows:

1. Data preprocessing and splitting
2. Initial model training without cross-validation
3. Cross-validation and hyperparameter tuning
4. Performance evaluation using metrics like accuracy, Hamming loss, and classification reports
5. Feature importance analysis

Links to Presentation and Paper

Presentation

Paper