An Effective Model for Diabetes Prediction Using Wrapper Technique

Abstract:

The eighth most common cause of elevated death rates in emerging nations is diabetes. When it is not appropriately monitored, it may develop into more severe, complex illnesses. Deficiency or ineffective insulin production is the major cause of diabetes and it has no permanent cure. Medical practitioners sometimes advise a healthy balanced routine, exercise, and early treatment. Machine learning algorithms have been helpful for the early prediction process. However, existing studies find it difficult to ascertain the most promising of them. This study utilized five (5) frequently used wrapper techniques (Ant Colony Optimization (ACO), Recursive Feature Extraction (RFE), Genetic Algorithm (GA), Firefly Algorithm (FA), and Particle Swam Optimization Algorithm (PSO) for building a Random Forest (RF) classifier. Nine (9) attributes and 10001 responses were used in training the model. Results showed that FA-RF outperformed the other model with an accuracy level of 99.91%, precision of 99.95%, specificity of 100%, and sensitivity of 98.96%. Results in all the models show confidence in the knowledge discovery from the dataset used.

Keywords: wrapper Technique, Diabetes prediction, Random Forest, ACO-RF, FA-RF, GA-RF