Diabetes mellitus is responsible for not only high health-care costs, but has also become a major cause of death prevailing over the past few decades, accounting for about 2, 23,000 deaths per year in India only due to it. Effective management and prediction of Diabetes can allow medical professionals to offer optimal treatment while lowering costs. Proper selections of machine learning algorithms are the parameters for implementation of such a decision support system. In this study, the different models were observed using various classifications, Decision tree techniques for each of the Data Reduction techniques to identify the best approach for diabetes prediction. In this paper, for detecting diabetes at an early stage over the instances of diabetic, non-diabetic and borderline patients we have chosen Gaussian Naive Bayes Algorithm in combination with Information Gain Attribute Evaluation, as we observed that this combination yields the best accuracy and performance.