Improving Accuracy for Classifying Selected Medical Datasets with Weighted Nearest Neighbors and Fuzzy Nearest Neighbors Algorithms

Abstract

Classification algorithms are very important for several fields such as data mining, machine learning, pattern recognition, and other data analysis applications. This work presents the weighted nearest neighbors and fuzzy k-nearest neighbors algorithms to classify chosen medical datasets. This involves several distance functions to calculate the difference between any two instances. Classification approaches based on K-nearest neighbors (KNN), weighted-KNN, frequency, class probability, and fuzzy K-nearest neighbors (fuzzy-KNN) are analyzed and discussed. Some measurable criteria are adopted to evaluate the performance of such algorithms. This includes classification accuracy, time, and confidence values. The algorithms will be tested using four different medical datasets. From the results, the fuzzy-KNN achieved the best accuracy compared to the other adopted algorithms. Following that are the weighted-KNN then the KNN. The longest classification time was for the fuzzy-KNN while the smallest time was for the KNN. The class confidence values of the fuzzy approach were promising. The fuzzy-KNN was also modified using fuzzy entropy. For the chosen datasets and w.r.t. KNN, the modified algorithms improved the classification accuracy. The improvements were up to 25%, 33%, and 38% for the weighted-KNN, fuzzy-KNN, and fuzzy Entropy respectively.