Deep Learning Models for Imbalanced Time Series Clinical Dataset

This repository contains experiments and implementations of various Deep Learning models aimed at improving performance on imbalanced time-series clinical data. The project explores data augmentation techniques and model optimization to enhance classification results in challenging imbalanced datasets.

Project Overview

The primary goal of this project is to apply and experiment with different deep learning methods to handle imbalanced time-series clinical data. Several techniques, including data augmentation with SMOTE, model optimization, and feature selection, were employed to maximize performance metrics such as AUC.

Key Features:

• 10 Deep Learning Methods: Experimentation with models like RNN (with Echo State Network), Transformer, and others.
• Imbalanced Data Handling: Utilized SMOTE and Borderline SMOTE to address class imbalance (49% imbalance).
• Feature Selection: Employed Random Forest Feature Ranking to identify key features in the dataset.
• Performance Metrics: Achieved a maximum AUC score of 0.95 through careful model tuning.

Methods and Techniques

1. Data Augmentation:

   • SMOTE (Synthetic Minority Over-sampling Technique): Generated synthetic samples for the minority class.
   • Borderline SMOTE: A more advanced version of SMOTE that focuses on difficult-to-classify data points near the decision boundary.

2. Deep Learning Models Tested:

   • RNN with Echo State Networks (ESN): A variant of recurrent neural networks that simplifies training by using a reservoir of randomly connected neurons.
   • Transformer: Applied transformer-based models to capture complex patterns in the time-series data.
   • Random Forest Feature Ranking: Used to reduce dimensionality and focus on the most important features in the dataset.

3. Model Tuning and Optimization:

   • Performed hyperparameter tuning to optimize the deep learning models using techniques such as grid search and random search.
   • Used TensorFlow for model implementation and training.

Results

• Max AUC: 0.95 achieved after applying data balancing techniques and optimizing the deep learning models.
• Improvement in Time-Series Performance: Significant improvement in classification accuracy, recall, and F1-score due to the use of SMOTE and advanced deep learning architectures.

Requirements

• Python 3.x
• TensorFlow: pip install tensorflow
• scikit-learn: pip install scikit-learn
• imbalanced-learn: pip install imbalanced-learn (for SMOTE and Borderline SMOTE)
• numpy, pandas, matplotlib

To install all dependencies at once, run:

pip install -r requirements.txt