Sooryak12 / Hospital-Department-Predictor-Sahyadri-

ML Model which predicts which department doctor you have to book based on your symptoms.

Hospital-Department-Predictor-Sahyadri

ML Model which predicts which department doctor you have to book based on your symptoms.

https://hospital-department-predictor.herokuapp.com/ -- model deployed here

• Hospital-Department-Predictor Model is deployed using flask as backend.
• Due to the unavailability of good datasets in healthcare sector.I have choosed dataset[https://www.kaggle.com/itachi9604/disease-symptom-description-dataset] which has 4000 datarows,but it has only 316 unique rows.
  
• the dataset mapped symptoms to diseases. I cleaned the data and mapped it to Hospital Departments.
  
• So I have extra 50-60 rows of symptoms with the help of Apollo Hospitals India [https://www.apollohospitals.com/patient-care/health-and-lifestyle/diseases-and-conditions/] and with the help of my friends studying doctorate.

Details:

• Refer either Models file or Models with Augmented Rows for preprocessing and Model Training.

• application.py files refers to the deployment of this model using flask as backend.

• raw datafile is training.csv

• Preprocessed data files are cleaned_data.csv and cleaned_data_with_4000_rows.csv

• best models are my_model1_4k and model4rf.sav

I have trained 3 Models with 350 unique rows:[Refer Models.ipynb file]

           1. Dense neural Network with 1 hidden Layer.<br />              
           2. Dense neural network with 2 hidden layer.<br />              
           3. Random Forest Classifier. ["Best Model in this section"] <br />
           
   *Note: Neural Networks showed bias to general physician Category as majority of rows were of them*

I have trained 3 Models with 4000 rows(Refer Models with Augmented Rows.ipynb file .350-370 rows were interpopulated to produce an unbiased model)

           1.Dense neural Network with 1 hidden Layer.["Best Model Overall"]<br /> 
           2.Dense neural network with 2 hidden layer.<br />
           3.Random Forest Classifier.<br />
           
        *Note :Neural Networks  provided excellent results with manual testing and this has been deployed in ngrok.*

• The jupyter notebook files are not documented well.
  

• Though My models produce excellent results,still they lack knowledge and make mistake in some simple cases.This is especially because of the poor dataset.
  

• This predictor was more of a data collection ->Test the Models -->Note the Mistakes by manual testing -->Data Collection -->Repeat Task for me.
  

• This signifies how important a good dataset is to a model
  

As we collaborate with hospitals to deploy our app we plan to accumulate data from them and build a good dataset.