Diabetic Prediction using Feature Selection based Random Forest and Fine Tuned K-Nearest Neighbor Classifier Algorithm-A Design Thinking Approach

Abstract

In low- and middle-income nations today, diabetes affects the majority of the population, according to a World Health organization (WHO) research. The WHO report suggested that 80% of the deaths would be due to the diabetes from 2016 to 2030. However, the current method continues to provide findings that are erroneous, which has a substantial negative impact on performance. To overcome the abovementioned issue, in this work, Random Forest (RF) algorithm and Fine tuned K-Nearest Neighbor (FKNN) classifier algorithm is proposed. Pre-processing, feature selection, and classification are the three primary stages of this project. Initially, preprocessing is performing for improving the final dataset results more accurately. Preprocessing is the process of cleaning the database into correct format. In order to choose more relevant and useful data from the dataset, the feature selection is then carried out utilizing the RF algorithm. It also minimizes the risk of over fitting with minimum features. Finally, diabetic prediction and classification is done by using FKNN classifier algorithm is used for categorizing items in the feature space based on training samples that are the most similar to the objects being classified. According to the experimental results, the suggested RF+FKNN method outperforms the current algorithms in accuracy, precision, recall, and f-measure.