AI-Disease-Prediction-System

AI-Disease Prediction System Project

This project focuses on the model of predictive diagnosis involving data munging, feature selection, model training and accuracy calculation. We try to introduce and validate methods that improve certain aspects of this process.

Assignment Problem Statement & Details

Problem Statement [3-PART]:

1. The data munging process introduces significant bias depending on majority of dataset by replacing with mean.
2. The feature selection process suffers from lack of adaptability by keeping a constant p-value significance limit of 0.05 without variance depending on the disease.
3. The datasets for diagnosis prediction must be general & not bias, with enough data for each class of features.

For data munging, we propose improvements through using the method of k-nearest-neighbors and extensively compare the results with data munging done by replacing with mean.

For the feature selection, we use the simulated annealing algorithm to set up an environment of varying p-value within a given upper bound and lower bound, defined initial temperature and cooling rate. Evaluation of the model has been done by testing on the training set rather than dividing the sets, although results from cross-validation have been provided at numerous places throughout the report. This has been supported by the fact that the 90/10 Train-test method gave stale accuracies for most p-values although they had different attribute sets and considerable differences when predicting on the training set. For data augmentation, to improve the diversity of the dataset - we have proposed a few data augmentation techniques such as adding noise, scaling etc.

Instructions for running the application:

1. Prerequisite: A local mysql server.
2. Open the DBConnection.java file in each of the two projects (BreastCancerPrediction & HeartDiseasePrediction).
3. Change the username and password in these two instances as per the credentials set in your system during your mysql installation. (For most mysql servers, the defaults are username: “root”, password: “”).
4. Run the BreastCancerPrediction.java and the HeartDiseasePrediction.java independently. Note: The main functions of each of the two programs do not run all the functions in sequence. You may uncomment the function calls to get the outputs of specific functionalities as needed.