Machine Learning to Diagnose Common Diseases Based on Symptoms

Abstract

It is often difficult for some people in rural India to diagnose a disease at an early stage due to lack of healthcare facilities and knowledge about its symptoms. Early diagnosis is the key for effective treatment of a disease and better living of the people. Providing sophisticated and accurate algorithms and techniques to overcome this issue will be revolutionary. Though hospitals in urban areas are using advanced technology for diagnosis and prognosis, proper diagnosis and prediction of diseases are a very hypercritical task. With the advancement of information technology and data sciences, disease diagnostic tools can be introduced to the general public, which can lower the burden of often depending on more expensive medical diagnostic technologies. Machine learning enables a system to learn from the previous data and takes decisions of its own. The programmer uses statistical techniques and feeds programs to the system using which it can store data and make decisions based on previous knowledge. The performance of the program increases with more and more training data. In this paper, we present a machine learning technique using Decision Tree Algorithm (DTL) to interconnect the symptoms and rearrange them and retrieve the most probable diagnosis. This technique allows the system to self-learn without using programming. This paper presents a system for diagnosis of common diseases, i.e., diabetes and heart diseases; by entering the symptoms into the system. Root node entropies are calculated for these two diseases by using the information gain for all the symptoms associated with these two diseases. The symptom values were categorized into three different levels, i.e., Low, Normal and High.