A Mathematical model for Predicting Nearest Neighbor in Diabetes Diagnosis

Diabetes has a significant impact on a substantial proportion of the global population. It is widely regarded as the most prevalent global ailment, as it impacts individuals of all ages and socioeconomic backgrounds. The integration of artificial intelligence into the field of medicine has facilitated the deduction of numerous diseases and has also aided in the anticipation and timely identification of various ailments, such as diabetes.

This research presents a novel classification algorithm that relies solely on mathematical computations to accurately predict the health status of patients, distinguishing between those with diabetes and those without. By doing these computations on a set of patient attributes, such as BMI, pregnancies, insulin level, etc., which are associated with diabetes, we can derive values that are utilized to forecast the patient's condition by comparing them with the closest categorized values.

The results of the proposed study demonstrate that our suggested algorithm PNN surpasses existing machine learning algorithms, including Decision Tree, Naïve Bayes, AdaBoost, and knn, in terms of accuracy.

The highest accuracy obtained by our proposed algorithm PNN is 83%, which is achieved when k= 17. That is higher than all the algorithms tested (AdaBoost 72%, Decision Tree 68%, Naïve Bayes 66%, KNN 78%)

Calculating the average of average of all features have increased the accuracy ratio.