Data Mining berbasis Nearest Neighbor dan Seleksi Fitur untuk Deteksi Kanker Payudara

Abstract

Detecting breast cancer in early stage is not straightforward. This happens because biopsy test requires time to determine whether the type is benign or malignant. Data mining algorithm has been widely used to automate diagnosis of a disease. One of popular algorithms is nearest neighbor based because of its simplicity and low computation. However, too many features can cause low accuracy in nearest neighbor based models. In this research, nearest neighbor based with feature selection is developed to detect breast cancer. Conventional k-Nearest Neighbor (KNN) and Multi Local Means k-Harmonic Nearest Neighbor have been chosen as nearest neighbor based models to experiment. The feature selection method used in this study is filter based, namely Correlation based, Information Gain, and ReliefF. The experimental result shows that the highest recall metric of MLM-KHNN and Information Gain is 94% with 5 features. In brief, MLM-KHNN algorithm with Information Gain can increase the recall of the prediction of breast cancer compared with the conventional K-NN algorithm and have been deployed into website using Streamlit such that the model can be used to detect breast cancer from chosen Wisconsin dataset features.