Modified K-Nearest Neighbour Using Proposed Similarity Fuzzy Measure for Missing Data Imputation on Medical Datasets (MKNNMBI)

Abstract

Early disease diagnosis is a burning problem in health sector, medical domain and disease management. During analysis, quality of the data can be achieved only if the data is complete. Missing values reduces the efficiency of data analysis task. Researchers proposed various imputation methods but always there was a need for a better imputation method. This paper objective is to propose a method for imputation using proposed similarity fuzzy measure through which we can impute missing values by finding k similar instances called as Modified k-Nearest Neighbour for imputation of missing data (MKNNMBI). The proposed imputation method outperformed when compared with other existing imputation methods MV EM, MV BPCA, MV Ignore, MV KMeans, MV FKMeans, MV KNN, MV MC, MV WKNNimpute, MV SVDimpute, MV SVMimpute, CBC-IM-FUZZY. These imputation methods were studied on different benchmark datasets and tested for performance on different classifiers like C4.5, SVM, kNN, NB and found that the proposed method leads to accurate imputation and improves the accuracy.