Insurance Claim Prediction Model

🎯 Project Overview

This project implements a Binary Classification machine learning model designed to predict whether an insurance policyholder will file a claim.

• Target Variable: 1 (Claim Filed) vs 0 (No Claim).

🧠 The Machine Learning Pipeline

Our robust, automated pipeline handles the end-to-end data journey:

1. Data Preprocessing: Cleans historical policy data and encodes categorical features to prepare it for the model.
2. Feature Scaling: Uses our ScalerFactory to standardize numerical features, ensuring all data points are treated equally by the algorithm.
3. Imbalanced Data Handling: Because insurance claims are rare events, the dataset is highly imbalanced. We utilize our SamplingStrategyFactory (leveraging imbalanced-learn) to synthesize or balance the training data, preventing the model from simply predicting "No Claim" every time.
4. Modeling: A configurable classifier (managed by the ModelFactory) learns the complex patterns that indicate a high likelihood of a future claim.

📊 Evaluation Metrics

Standard accuracy can be highly misleading for imbalanced datasets. Therefore, our ModelEvaluator heavily focuses on metrics that provide a true picture of predictive performance:

• F1-Score: The F1-Score is the harmonic mean of precision and recall. It provides a balanced measure of a model's performance that is especially useful for imbalanced datasets.
• ROC AUC: Receiver Operating Characteristic - Area Under Curve (ROC-AUC) evaluates the model's overall performance across different thresholds. It measures the model's ability to distinguish between classes.
• False Negative Rate (FNR): The percentage of actual claims that our model incorrectly predicted as "No Claim." In the insurance domain, a False Negative is typically the most expensive mistake (failing to reserve capital for a valid claim), making this our primary optimization target.

📈 Expected Outputs

Running the pipeline (via python src/main.py or ./run.sh) generates the following artifacts:

• Trained Model File: A serialized model ready for inference on new data.
• Console Logs: Real-time evaluation scores (F1, ROC AUC, FNR) outputted to your terminal.
• Visualizations: Automatically generated plots saved in the output directory, including:
  • Confusion Matrix: A visual breakdown of true vs. false predictions.
  • ROC Curve: A graphical representation of the model's diagnostic ability.