DiabeDetect: A Novel Decision Tree-Based Approach for Early Prognosis of Diabetes

Diabetes is one of the most serious threats for human throughout the world and currently this is owne of the most threatening cases in low and low-middle-income countries. However, the early prognosis can significantly reduce the threat to a minimal level through proper care-coordination schemes. This paper aims at finding the likelihood probability of diabetes using different machine learning techniques along with a novel approach aided by Decision Tree to improve the performance. Diabetes patients data were collected from the early stage diabetes risk prediction dataset of the UCI machine learning Repository. This dataset is consisted of 17 attributes in which we applied Logistic Regression, Naive Bayes, variants of Support Vector Machine (SVM), and our novel decision tree aided approach and calculated their prediction accuracy. An efficient exploratory data analysis coupled with the attribute correlation method helped us to develop the novel approach that improved the accuracy by deducting some lower ranked misguided attributes. After a careful selection of upper ranked attributes, we found a much-improved accuracy rate of 92% for our Novel approach that outperforms the predecessor approaches.