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.
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