AN EARLY DISEASE PREDICTION MODEL FOR DIABETES PATIENTS WITH MACHINE LEARNING APPROACH

Abstract

Now-a-days most of the people are affected by diabetes disease and this leads to create more problems such as cardiovascular disease and other health issues. Several methods exist to identify the diabetes disease in an earlier stage but they are failing to predict the diabetes disease in an accurate manner and also consume more time. To overcome these issues the effective diabetes detection techniques are proposed. Initially, the amount of training data is increased with sliding window concept. Then imbalance dataset are balanced using the adaptive sampling technique. Further the diabetic recognition process is improved by applying the Intensity Weighted Firefly Optimization firefly techniques (IWFO). This method selects the features based on the correlation between the features which reduces the irrelevant features involved in the diabetic recognition process. Then PCA based feature transformation technique is applied to handle the different type of feature. The selected features are classified by Hybrid random forest into two classes namely normal and diabetes. The network predicts the disease based on the relevance between each feature which helps to maximize the prediction accuracy. Further the efficiency of diabetes prediction system is enhanced by applying the present model on different dataset. At last, efficiency of the system is evaluated using Java based simulation results. From the analysis, it is clear that the proposed model predicts the diabetes disease with the minimum misclassification error rate and maximum accuracy when compared to the other methods.