Diabetes Dataset

Dataset Overview

Source

• Derived from the scikit‑learn library, containing data for 442 diabetes patients.

Content

• Features intended for predicting disease progression.

Methodology

Data Analysis

• Data preparation, visualization, and statistical analysis were performed to understand the dataset.

Model Development

• Implemented Multi‑Layer Perceptron (MLP), Linear Regression, and Deep Neural Network (DNN) models using PyTorch.

Training & Validation

• The dataset was split into training (80 %) and validation (20 %) sets for model evaluation.

Results & Validation

Model Evaluation

• Models were trained with the Adam optimizer and mean‑squared‑error loss. Validation losses were recorded to assess performance.
• Training and validation loss curves were plotted to visualize learning dynamics and detect overfitting.

Best Model

• Based on average validation loss, the Linear Regression model performed best.

Performance Metrics

Loss Metrics

• Detailed loss figures and summaries were provided to understand model behavior. Loss histories show how training and validation loss evolve over time.

Model Selection

• The model with the lowest average validation loss (Linear Regression) was selected as the optimal choice.

Conclusion

• The project successfully implemented and compared various machine‑learning models for predicting diabetes progression. The Linear Regression model achieved the best validation loss, demonstrating the effectiveness of the chosen architecture and hyperparameters.