Project details:
This project builds a regression model that predicts the recovery rate of gold from raw ore during extraction and refinement. The analysis covers process parameters, intermediate outputs and final concentrate characteristics. The final model helps mining operations assess ore quality early and avoid launching unprofitable production cycles.
Description
Business Context & Problem
Gold extraction involves several processing stages, each affecting how much metal can ultimately be recovered from the ore. Running a batch with poor recovery leads to financial loss, so being able to estimate efficiency before full processing is critical. This project focuses on predicting the recovery rate to support better operational planning and reduce risk.
Data & Analytical Approach
The dataset included features from multiple stages of ore processing: rougher feed composition, flotation outputs, cleaner stages and final concentrate measurements. After cleaning and preparing the data, exploratory analysis helped understand how different parameters influence recovery. Several engineered features were added to capture ratios and process-stage relationships. Data distributions were examined carefully because industrial data often contains noise and irregularities.
Statistical / ML Analysis
Using Scikit-learn, multiple regression algorithms were tested to predict both rougher recovery and final recovery. Performance was evaluated with appropriate metrics (such as SMAPE) that reflect the nature of industrial regression problems. Model tuning and cross-validation ensured stable results across samples. The best-performing model provided reliable estimates that generalise beyond the training data.
Key Insights & Final Recommendations
The modelling showed that several early-stage parameters strongly influence final recovery, meaning that ore quality can be assessed before running the full processing chain. The final model can be integrated into a production workflow to flag low-quality batches and help engineers avoid unprofitable operations.
Overall, the project provides a practical ML solution for improving efficiency and decision-making in gold extraction.