Predictive Analysis and Prognostic Approach of Diabetes Prediction with Machine Learning Techniques

Medical experts indulge in numerous strategies for efficient and predictive measures to model the health status of patients and formulate the patterns that are formed in test results. Most patients would dream of their betterments of their health conditions and thus preventing the progression of any disease. When diabetics is considered in the model, or highly intervening methodology would be required for pre-diabetic individuals. Hidden Markov models have been modified into variant models to derive predictions that accurately produce expected results by investigating patterns of clinical observations from a detailed sample of patient's dataset. There are yet unanswered and concerning challenges to derive an absolute model for predicting diabetes. The datasets from which the patterns are derived from, still holds levels of in completeness, irregularity and obvious clinical interventions during the diagnosis. The Electronic Medical Records are not furnished with all requisite information in all conditions and scenarios. Due to these irregularities prediction has become highly challenging and there is increase in misclassification rate. Newton's Divide Difference Method (NDDM) is a conventional model for filling the irregularity in electronic datasets through divided differences. The classical approach considers a polynomial approximation approach, thus leading to Runge Phenomenon. If the interval between data fields id higher, severity of finding the irregularities is even higher. By using this type of technique it helps in improving the accuracy thereby bringing in high level prediction without any error and misclassification. In this technique proposed, a novel approximation technique is implemented using the Euclidean distance parameter over the NDDM approximation to predict the outcomes or risk of Type 2 Diabetes Mellitus among patients. Real world entities in CPCSSN are considered for this study and proposed method is tested. The proposed method filled the irregularity in the data components of EMR with better approximations and the quality of prediction has improved significantly.