HealLink - Machine Learning Model

This repository contains experiments and evaluations for a mental health classification model, aiming to classify user text into specific mental health categories.

• Introduction
• Dataset Details
• Model Architecture
• Requirements
• Usage
• Results
• Future Work
• Acknowledgments

Introduction

The goal of this project is to create a robust NLP model capable of classifying text into specific mental health categories, such as depression, anxiety, and suicidal thoughts. The focus is on leveraging pre-trained BERT models and improving performance through techniques like focal loss and dropout regularization.

Dataset Details

• Source: kaggle dataset
• Preprocessing:
  • Text tokenization using BERT's tokenizer.
  • Standardization: Lowercasing.
• Imbalance Handling: Addressed class imbalance using weighted loss functions.

Model Architectures

1. BERT Base (L-12, H-768, A-12)
   • Pretrained weights: bert-en-uncased-l-12-h-768-a-12

Requirements

1. Anaconda or Miniconda

Usage

1. Clone the repository:

   git clone https://github.com/HealLink/ML-Model.git
   cd ML-Model

2. Create and activate a Conda environment

   conda create -n model-env python=3.11.10 -y
   conda activate model-env
   

3. Install dependencies

   pip install -r final_requirements.txt
   

4. Run notebook_final.ipynb inside the notebooks subdirectory

Results

• The current best model achieved:
  • Epoch: 2
  • Train loss: 0.12276183813810349
  • Val loss: 0.082574762403965
  • Train MCC: 0.7533358335494995
  • Val MCC: 0.7541943192481995
  • Train Accuracy: 0.8077240586280823
  • Val Accuracy: 0.8009008765220642
• Confusion Matrix (Test set):

Acknowledgements

• Thanks to Allah Subhanahu Wa Ta'ala for all his grace and favor, so that this project can be completed properly.
• Thanks to TensorFlower for creating TensorFlow Framework.
• Thanks to Google for creating BERT as base model.
• Thanks to Mr. Andrew Ng and Mr. Laurence in Coursera for teaching ML.
• Thanks to Bangkit Teams for this learning opportunity.
• Thanks to Suchintika Sarkar for compiling and cleaning the dataset.