Machine Learning

Heavy Equipment Price Prediction

End-to-end ML pipeline on 138,701 auction records — improved RMSLE from a 0.2045 baseline to 0.18732 with a LightGBM + CatBoost ensemble.

Role: Solo — IITM Machine Learning Practice (Kaggle competition)
Timeline: Aug 2026 — Oct 2026

Problem

Predict the auction sale price of heavy equipment from a messy, high-cardinality dataset (138,701 rows × 50 columns) for the IIT Madras Machine Learning Practice Kaggle competition.

Approach

Thorough EDA and data cleaning: missing-value strategy, unit normalisation and outlier handling
Feature engineering from dates, model descriptors and usage features
Baselines with linear and tree models, then tuned LightGBM and CatBoost
Weighted ensemble validated with time-aware splits

Result

RMSLE improved from 0.2045 → 0.18732, with a reproducible notebook-to-submission pipeline.