Kaggle Competitions

A collection of my Kaggle competition solutions, notebooks, and learnings

About
Competitions
Repository Structure
License

About

This repository contains my solutions and approaches to various Kaggle competitions. Each competition folder includes:

Jupyter notebooks with complete solution code
Detailed README explaining the approach
Model architectures and training configurations
Results and performance metrics

Competitions

Competition	Description	Result	Approach
NeurIPS 2025 - Google Code Golf Championship	Shortest code wins! Google Code Golf at NeurIPS 2025 - Write the most concise programs to solve challenging tasks.	Top 24% (Score: 264)	Deep pattern analysis, hypothesis-driven solution development, code golf optimization with batch compression & verification optimizer for 400 ARC-AGI tasks
MABe Challenge - Social Action Recognition in Mice	Detect unique behaviors from pose estimates of mice across 400+ hours of footage	Top 11% (Score: 145)	LightGBM + XGBoost + CatBoost Ensemble with separate GBDT models per `body_parts_tracked` config, extensive feature engineering (distance, velocity, angle, wavelet features), and GroupKFold CV
Fake or Real: The Impostor Hunt in Texts	ESA's "Secure Your AI" series - Identify real vs. fake LLM-generated documents	Top 12% (Score: 114)	5-Head DeBERTa-v3-Large Ensemble (Mean Pool, Max Pool, CLS Token, Attention Pool, Concat Pool) with Optuna-optimized voting, achieving ~0.925 CV accuracy
ARC Prize 2025	Advanced Reasoning Challenge - Solve ARC tasks with ML/AI	Top 29% (Score: 413)	Multi-stage pipeline combining DSL-based symbolic program search (DFS), CNN-based neural-guided program synthesis (SketchPredictor), and adaptive task classification with fallback predictions
Predicting Student Test Scores	Predicting standardized test outcomes for students	Top 12% (Score: 478)	Regression with feature engineering, model selection & ensemble strategy, evaluated on RMSE (Playground Series S6E1)
Predicting Loan Payback	Loan default risk prediction	Top 11% (Score: 404)	12-stage feature engineering pipeline (risk, affordability, ratios, bins, interactions), multi-model ensemble with configurable weights and folds, ROC-AUC optimized
Predicting Road Accident Risk	Predict road accident risk on different road types	Top 1% (Score: 29)	XGBoost + LightGBM + CatBoost weighted blending with interaction features (speed_curvature, lanes_speed, curvature_squared), weather+lighting risk flags, and RandomizedSearchCV tuning
Binary Classification with a Bank Dataset	Predict bank customer responses for targeted marketing	Top 5% (Score: 151)	CatBoost classifier with native categorical handling, 10-fold Stratified CV, GPU training, early stopping, trained on combined competition + original Bank Marketing dataset
Predicting the Beats-per-Minute of Songs	Predict the tempo (BPM) of songs using audio/music features	Top 2% (Score: 38)	LightGBM + XGBoost + CatBoost + Random Forest ensemble with weighted blending & Ridge meta-model stacking, rhythm/duration/intensity feature engineering

Repository Structure

Kaggle_Competitions/
├── README.md                                           # This file
├── LICENSE                                             # MIT License
│
├── NeurIPS 2025 - Google Code Golf Championship/       # NeurIPS 2025 Code Golf
│   ├── README.md                                       # Detailed solution writeup
│   └── google-code-golf-championship-101.ipynb         # Main notebook
│
├── MABe Challenge - Social Action Recognition in Mice/ # Mouse Behavior Competition
│   ├── README.md                                       # Detailed solution writeup
│   └── mabe-challenge.ipynb                            # Main notebook
│
├── Fake or Real The Impostor Hunt in Texts/            # ESA NLP Competition
│   ├── README.md                                       # Detailed solution writeup
│   └── fake-or-real-the-impostor-hunt-in-texts.ipynb   # Main notebook
│
├── ARC Prize 2025/                                     # ARC Prize 2025
│   ├── README.md                                       # Detailed solution writeup
│   └── arc-prize-2025-comprehensive-solution.ipynb     # Main notebook
│
├── Predicting Student Test Scores/                     # Student Score Prediction
│   ├── README.md                                       # Detailed solution writeup
│   └── student_test_scores.ipynb                       # Main notebook
│
├── Predicting Loan Payback/                            # Loan Payback Prediction
│   ├── README.md                                       # Detailed solution writeup
│   └── predicting-loan-payback.ipynb                   # Main notebook
│
├── Predicting Road Accident Risk/                      # Road Accident Risk Prediction
│   ├── README.md                                       # Detailed solution writeup
│   └── predicting-road-accident-risk-101.ipynb         # Main notebook
│
├── Binary Classification with a Bank Dataset/          # Bank Term Deposit Prediction
│   ├── README.md                                       # Detailed solution writeup
│   └── bank-term-deposit-prediction.ipynb              # Main notebook
│
└── Predicting the Beats-per-Minute of Songs/           # BPM Prediction
    ├── README.md                                       # Detailed solution writeup
    └── predicting-the-beats-per-minute-of-songs-101.ipynb   # Main notebook