Improve Model Performance on Imbalanced Data via SMOTE and Scaling

[Arin-12]

Description

The current machine learning models in this project (specifically the Random Forest implementation) show a significant disparity between Accuracy (~90%) and Recall (~50%). This indicates a class imbalance issue where the model struggles to identify the minority class effectively.

Proposed Improvement

I have implemented a pipeline that uses:

1. SMOTE (Synthetic Minority Over-sampling Technique) to balance the training set.
2. StandardScaler to normalize feature distributions.

Results

These changes result in a much more balanced and reliable model:

• Recall: Improved from 0.50 to 0.63 (+26% gain)
• F1-Score: Improved from 0.59 to 0.64
• Precision: Maintained at a healthy 0.65

Checklist

• Code follows Python 3.x standards.
• Descriptive comments included for all new logic.
• MIT License added to the top of the file.

I have the code ready and would like to be assigned to this issue to submit a Pull Request!