Data Science

Overview

The Steel Waste Prediction System is an advanced machine learning application designed to predict steel waste percentage in construction projects and provide actionable insights to reduce material waste, costs, and environmental impact.

Built using state-of-the-art machine learning algorithms, the system analyzes 16 key project parameters to deliver accurate predictions with comprehensive explainability and reliability indicators.

Technology Stack

Machine Learning
  • Gradient Boosting Regressor (Best Model)
  • 10 different ML models tested
  • 10-fold cross-validation
  • 93% accuracy (R² Score)
Technology
  • Python & Flask
  • Scikit-learn
  • XGBoost
  • SHAP for explainability

Key Features

Model Explainability

SHAP-based explanations show which factors contribute most to waste predictions, providing transparency and actionable insights.

Reliability Indicators

Confidence scores (High/Medium/Low) indicate prediction reliability based on input data similarity to training data.

Cost Analysis

Calculate financial impact of predicted waste and potential savings from waste reduction strategies.

CO₂ Impact

Estimate environmental footprint and potential CO₂ reductions from optimizing material usage.

Model Performance

Metric Value
Test MAE0.73%
Test RMSE0.92%
Test R²0.93
Test MAPE10.88%
CV R² (Mean ± Std)0.92 ± 0.01

Input Parameters

The system analyzes 16 key project parameters:

MATERIAL PARAMETERS
  • • Reinforcement Ratio
  • • Number of Unique Lengths
  • • Stock Length Policy
  • • Cutting Optimization
MANAGEMENT & CONTROL
  • • Supervision Index
  • • Material Control Level
  • • Storage Handling Index
  • • BIM Integration Level
  • • Offcut Reuse Policy
PROJECT DYNAMICS
  • • Design Revisions
  • • Change Orders
  • • Contract Type
  • • Lead Time
  • • Order Frequency
  • • Project Phase