README.md

Predicting Electricity Usage in Steel Industry

Project Overview

This project focuses on predicting electricity consumption (Usage_kWh) in the steel industry using historical operational data. We explore both linear regression and Random Forest regression to handle skewed data and capture nonlinear relationships in the dataset.

Dataset

• Source: steel_industry_data.csv
• Key columns:
  • Usage_kWh: Electricity usage (target)
  • Lagging_Current_Reactive.Power_kVarh
  • Leading_Current_Reactive_Power_kVarh
  • Lagging_Current_Power_Factor
  • Leading_Current_Power_Factor
  • NSM: Number of shifts or operational measure
  • CO2(tCO2): CO2 emissions

Methodology

1. Exploratory Data Analysis (EDA):
   • Checked distributions and skewness.
   • Visualized histograms and boxplots for key variables.
2. Data Preprocessing:
   • Log transformation applied to reduce right-skew in Usage_kWh.
3. Modeling:
   • Linear Regression with and without log-transformed target.
   • Random Forest Regression with hyperparameter tuning (n_estimators, n_jobs=-1 for parallelization).
4. Evaluation:
   • Metrics: RMSE, MSE, R².
   • Residual analysis and distribution comparison.
5. Visualization:
   • Predicted vs actual plots, residual plots, log-transform effects.

Results

• Linear Regression (log-transformed target): Captured central tendency but underestimates high usage values.
• Random Forest Regression:
  • Captures nonlinearity and extreme values better.
  • RMSE significantly lower and distribution of predictions closely matches true data.
  • Handles skewed data without transformation.

Key Features / Skills Demonstrated

• Python, pandas, numpy, matplotlib, scikit-learn.
• Handling skewed distributions and feature analysis.
• Model comparison and evaluation metrics.
• Residual analysis and visualization.
• Feature importance interpretation using Random Forest.

Folder Structure