Diabetes has become one of the major health concerns for the modern day population. This can be attributed to a number of factors such as unhealthy lifestyle, meager diet, genetics, obesity, etc. The rapid growth in the number of diabetic patients urges the requirement for a state-of-the-art healthcare against such diseases. Early prediction of such diseases can be very useful for mitigating the risks associated with such diseases. In this context, this research proposes an indigenous efficient diagnostic tool for the detection of diabetes. The proposed methodology comprises two phases: Phase-I deals with collection of Pima Indian Diabetes Dataset from the UCI machine learning repository databases and Localized Diabetes Dataset from Bombay Medical Hall, Upper Bazar Ranchi, Jharkhand, India. In Phase-II, the acquired datasets are processed and analyzed using two different approaches. The first approach entails classification through Logistic Regression, K-Nearest Neighbor, ID3 DT, C4.5 DT, and Naive Bayes. The second approach employs PCA and PSO algorithms for feature reduction prior to the classification of the dataset using the methods used in the first approach. A comparative analysis is performed between the various approaches used in this manuscript. Results obtained clearly depict the efficiency of the proposed approach over the traditional classification approach in terms of less computation time and increased accuracy. The proposed approach has the potential to be applied for effective and early diagnosis of other medical diseases as well.
|Journal||Data powered by TypesetNetwork Modeling Analysis in Health Informatics and Bioinformatics|
|Publisher||Data powered by TypesetSpringer Science and Business Media LLC|