An Improved Method for Predicting Diabetes Mellitus Using Adaptive Neuro-Fuzzy Inference System

Chronic diseases are considered the major cause of death and disability worldwide. Diabetes is a chronic disease that occurs when the pancreas cannot produce enough insulin or when the body does not use the insulin effectively. According to World Health Organization (WHO), in the year 2019 alone, diabetes was the direct cause of about 1.5 million deaths. Since diabetes mellitus type 2 has become one of the major causes of premature diseases such as heart disease and kidney disease leading to death in many countries, it is important that an expert system be implemented and used in the diagnosis of this condition. Although several systems have been proposed and designed to diagnose diabetes mellitus type 2, the accuracy of different data mining and machine learning techniques is still not very high. Also, in cases where the accuracy of the prediction was high, it was discovered that very few input metrics were considered, which can actually be affected in a real life scenario. In this paper, a fuzzy logic based expert system model for diagnosing diabetes mellitus type 2 was developed. The developed model was evaluated alongside a similar fuzzy expert system. Several experiments were carried out to analyze the performance of the two models. Results showed that the model needed about 25 iterations to attain the global minimum, while the developed model needed 15 iterations, thus consuming less computation resources. Results also showed that the developed model outperformed the other model since it employed an augmented dataset. When tested with test dataset from the locally generated dataset, the developed fuzzy expert gave a prediction accuracy of 97%, with a specificity of 95%, a sensitivity of 94%, and a precision of 93% when compared to the other system that had a corresponding accuracy of 89%, specificity of 86%, sensitivity of 87% and a precision of 80%. This helps to establish the fact that there is a need to incorporate datasets that are local or unique to a group of persons or region so as to improve the accuracy of the developed model.