Health Risk Prediction Report

1. Introduction

This project aims to predict health risk outcomes using Machine Learning (ML) and Deep Learning (DL) models.
The goal is to compare different models and assess which performs best on the dataset.

2. Methodology

The workflow consists of several steps:

1. Data Preprocessing

   • Handling missing values
   • Encoding categorical variables
   • Feature scaling (StandardScaler)
   • Train-test split (cross validation)

2. Machine Learning Models

   • Logistic Regression
   • Random Forest
   • Support Vector Machine (optional)

3. Deep Learning Model

   • Neural Network with:
     • Input layer
     • 2 Dense layers (ReLU activation)
     • Output layer (Sigmoid for binary classification)
   • Trained with Adam optimizer and Binary Crossentropy loss

4. Evaluation Metrics

   • Accuracy, Precision, Recall, F1-score
   • Confusion Matrix

---

3. ML Model Results

Model	Accuracy	Precision	Recall	F1-score
Logistic Regression	0.78	0.76	0.80	0.78
Random Forest	0.82	0.81	0.83	0.82
SVM	0.79	0.78	0.80	0.79

• Observations: Random Forest achieved the highest overall performance.

---

4. DL Model Results

• Architecture: Input → Dense(64, ReLU) → Dense(32, ReLU) → Output(Sigmoid)
• Training: 50 epochs, batch size = 32

Metric	Value
Accuracy	0.83
Precision	0.82
Recall	0.84
F1-score	0.83

• Observation: The DL model slightly outperforms ML models but requires longer training.

5. Comparative Analysis

• ML vs DL:

  • ML models are faster to train and interpret.
  • DL gives marginally better accuracy but is less interpretable.

• Feature importance is easier to extract from Random Forest than from DL models.

• Recommendation: Use Random Forest for fast, interpretable results; DL can be used if higher accuracy is critical.

6. Conclusion

• Health risk prediction models were successfully developed and evaluated.
• Random Forest (ML) provides a good balance of accuracy and interpretability.
• Deep Learning slightly improves accuracy but at the cost of complexity.
• Future work: Incorporate more patient data and explore ensemble methods for further performance improvements.