Risk prediction analysis for classifying type 2 diabetes occurrence using local dataset

The steep rise of cases pertaining to Diabetes Mellitus (DM) condition among global population has encouraged extensive researches on DM, which led to exhaustive accumulation of data related to DM. In this case, data mining and machine learning applications prove to be a powerful tool in transforming data into a meaningful knowledge. Several machine learning tools has shown great promise in diabetes classification. However, challenges remain in obtaining an accurate model suitable for real world application. Most disease risk-prediction modelling are found to be specific to a local population. Besides that, real world data are likely to be complex, incomplete and unorganized making it a challenge to develop models around it. This research aims to develop a robust prediction model for classification of type 2 diabetes mellitus (T2DM), with the interest of a Malaysian population, using several well-known machine learning algorithm such as Decision Tree, Support Vector Machine and Naïve Bayers. In order to achieve this, several data pre-processing method is implemented to improve the model performance. The models utilize local based datasets obtain from IIUM medical centre records. Besides that, each models is validated using split and 10 cross fold method. Ultimately, the performance of each model is evaluated and compare based on several statistical metrics that measures the accuracy, precision, sensitivity and efficiency. The final result shows that Random forest model provides the best overall prediction performance in terms of accuracy (0.87), sensitivity (0.9), specificity (0.8), precision (0.9), F1-score (0.9) and AUC value (0.93) (Normal)