Smart Healthcare: Adding Rules to Machine Learning Models to Diagnose Diseases

Abstract

Machine learning tools have the potential to outperform humans in many different areas. One such area is healthcare. One area of healthcare where machine learning can be used is in aiding doctors with diagnosing patients. Decision trees and neural networks can help eliminate human error and have the potential to become better than doctors at diagnosing diseases. One problem is that computers don’t have the expertise that doctors do. The purpose of this project is to see if adding medical "rules" increases the accuracy of decision trees and neural networks during disease prediction. Five different medical classification datasets were used to train neural networks and decision trees. Decision trees were trained with multiple different combinations of rules found online. For neural networks, both rules found online, and decision tree produced rules were added to different layers. Results from this thesis revealed that rules increased the accuracy of both decision tree and neural network models for 4 out of the 5 datasets used. The decision tree prediction accuracy increased by an average of 6.15% and the neural network prediction accuracy increased by an average of 2.18%.