Mental_Stability_TransferLearning

A Novel Transfer Learning Approach for Mental Stability Classification from Voice Signal

This repository contains the implementation of the Transfer Learning and Data Augmentation methodology for classifying mental stability using voice signals. The research focuses on utilizing Convolutional Neural Networks (CNNs) and spectrogram analysis to detect mental health conditions.

📄 Research Paper

Title: A Novel Transfer Learning Approach for Mental Stability Classification from Voice Signal
Authors: Rafiul Islam, Dr. Md. Taimur Ahad, Bo Song, Yan Li
Status: Under review

📊 Project Overview

Mental health diagnostics often face challenges due to subjective assessments and limited resources. This study proposes a voice-based diagnostic approach that employs transfer learning approach to classify mental stability using CNN architectures like VGG16, InceptionV3, and DenseNet121 with data augmentation for improved classification accuracy..

Key Highlights:

Dataset:
- Collected Data: 85 voice recordings categorized as stable and unstable.
- Preprocessing: Spectrograms generated using Short-Time Fourier Transform (STFT).
Model Selection:
- CNN Architectures:
  - VGG16: Captures spatial hierarchies for fine-grained spectrogram details.
  - InceptionV3: Handles multi-scale feature extraction.
  - DenseNet121: Promotes feature reuse and achieves the best performance.
- Performance: DenseNet121 achieved 94% accuracy and an AUC of 0.99 that outperform other models.
Data Augmentation:
- Techniques: SpecAugment, Gaussian noise, random erasing.
- Improved model generalization on diverse audio conditions.
Transfer Learning:
- Pre-trained models on augmented data.
- Fine-tuned on non-augmented data for task-specific learning.

🚀 Getting Started

1. Clone the Repository

git clone https://github.com/rafi0020/Mental_Stability_TransferLearning.git
cd Mental_Stability_TransferLearning

2. Install Dependencies

pip install -U image-classifiers efficientnet

3. Prepare the Dataset

Download the dataset from Mendeley Repository (https://data.mendeley.com/datasets/s5j25b5tjk/1).
Store the data in the data/ folder.

4. Run the Notebook

Open Aug_TransferLearning_Code.ipynb in Jupyter Notebook or Google Colab.

📊 Results

Comparison Graph:

DenseNet121 outperform other models.

Key Visualizations:

Training/validation accuracy and loss curves
Confusion matrix
ROC curves

🤝 Collaboration Contributions are welcome! Feel free to open an issue or submit a pull request.

📫 Contact

Email: rafiulislam1921@gmail.com

Let me know if you’d like to proceed with implementing this structure, or if you want specific adjustments to the README.md. I can also help with creating scripts or setting up the repository locally!