Steel Energy Consumption Predictor

Comprehensive machine learning pipeline for predicting energy consumption (Usage_kWh) in the steel industry using temporal patterns, power metrics, and advanced feature engineering. Trained on 35,041 records spanning January-December 2018 at 15-minute intervals.

Features 79 engineered features from 11 originals. Benchmarks 9 algorithms including XGBoost and LightGBM. Best models achieve R² of 0.95-0.98.

Key Features

• 79 engineered features from 11 original inputs
• 9 ML algorithms benchmarked (Linear to Neural Network)
• XGBoost/LightGBM achieving R² 0.95-0.98
• Temporal features with cyclical encoding
• Lag features at 8 different intervals
• Rolling window statistics (1hr to 6hr)
• Hyperparameter tuning via RandomizedSearchCV
• Production prediction script included
• 6 Jupyter notebooks for full pipeline