Analysis And Comparison Of Diabetic Prediction Using Medium KNN Classifier And Cosine KNN Classifier

Aim: The goal of the study is to find out the presence of diabetes using the Medium K-NN (K- Nearest Neighbour) and Cosine K-NN (K- Nearest Neighbour) algorithm and comparing the accuracy, specificity and sensitivity. Materials and Methods: A compilation of information from Kaggle's website was used in this research. The samples were regarded as (N=25) for Medium KNN and (N= 25) Cosine KNN according to clinicalc.com, total sample size calculation was performed by keeping alpha error-threshold value 0.05, enrollment ratio is 0:1, 95% confidence interval and 80% power. The accuracy, specificity, sensitivity was calculated by using Matlab software. Results: The accuracy (%), specificity (%) and sensitivity (%) is compared using SPSS software using independent sample t tests. There is a statistically insignificant difference, p=0.219, p>0.05 with accuracy (47.2%), p=0.067, p>0.05 with specificity (34.54%) and p=<0.01, p<0.05 with sensitivity (48.36%) and demonstrated a better outcome in comparison to Cosine KNN accuracy (44.12%), specificity (39.28%) and sensitivity (45.64%). Conclusion: Medium KNN appears to give better accuracy, specificity and sensitivity than Medium KNN to predict diabetes.