Text-to-Disease-Classification

Formulated a large Language Model (LLM) by fine tuning pre-trained BERT transformer ,that can accurately classify disease name based on short description of symptoms.

Deployed this system as a Flask-based web application, ensuring seamless accessibility and user-friendly interaction.

Installation

Create a conda environment using environment file , activate that environment and the run the app.py file. if run sucessfully, you should be able to access the web app at localhost.

Example:-

# clone repo
git clone git@github.com:0x1h0b/Text-to-Disease-Classification.git

# create environment
conda env create --name <envname> --file=environments.yml

# activate env
conda activate <envname>

# run app.py
python app.py

Data

Used below kaggle dataset to fine tune the BERT model

Kaggle:// disease-symptom-description-dataset

notes :-

-  Data is equally Balanced, 120 symptom lists/entries for each 43 disease.
-  all disease have corresponding description and precautions
-  visualised the count plot of individual symptom.
    -  type of symptoms are more balanced and diverse in symptom1 and symptom2 feature column.
-  removed all chars other then aplha-numeric. removed extra spaces. 

Training & Evaluation

-  Used OOP by defining custom Dataset and Model class to load dataset and fine tune BERT model.
-  Used K fold cross Validation for Training and evaluation of model.
-  Trained the model for 3 epochs for 5 folds

notes :-

- FOLD 1 , Epoch: 2 ,loss : 3.163252592086792 , accuracy :0.991869918699187
- FOLD 2 , Epoch: 2 ,loss : 3.164623737335205 , accuracy :0.9956808943089431
- FOLD 3 , Epoch: 2 ,loss : 3.1687734127044678 , accuracy :0.9852642276422764
- FOLD 4 , Epoch: 2 ,loss : 3.1720407009124756 , accuracy :0.9933943089430894
- FOLD 5 , Epoch: 2 ,loss : 3.1956565380096436 , accuracy :0.9784044715447154

Final Results

when you enter the top 7 symptoms in the flask web app , it will display the top 5 disease based on the symptoms.

Here is the sample result from flask ,