Improving Classification Performance of K-nearest Neighbour by Hybrid Clustering and Feature Selection for Non-communicable Disease Prediction

Non-communicable Disease (NCDs) is the high mortality rate in worldwide likely diabetes mellitus, cardiovascular diseases, liver and cancers. NCDs prediction model have problems such as redundancy data, missing data, noisy class and irrelevant attribute. This paper proposes a novel NCDs prediction model to improve accuracy. Our model comprises k-means as clustering technique, Weight by SVM as feature selection technique and k-nearest neighbour as classifier technique. The result shows that k-means + weight by SVM + k-nn improved the classification accuracy on most of all NCDs dataset (accuracy; AUC), likely Pima Indian Dataset (96.82; 0.982), Breast Cancer Diagnosis Dataset (97.36; 0.997), Breast Cancer Biopsy Dataset (96.85; 0.994), Colon Cancer (99.41; 1.000), ECG (97.80; 1.000), Liver Disorder (97.97; 0.998).