Optimasi Kinerja Random Forest pada Klasifikasi Data Microarray Menggunakan Minimum Redundancy Maximum Relevance

Abstract

Cancer is a disease that can spread to other cells/tissues in the patient's body and its growth cannot be controlled. In Indonesia, the prevalence of cancer in the 2018 Riskesdes data was 1.79 per 1,000 population with cancer. Due to the high prevalence of these cancers, it is necessary to detect cancer early. One way to detect cancer is by gene expression using microarray technology, which can monitor thousands of gene expressions simultaneously in one experiment. However, microarray data have huge dimensions, so it is necessary to reduce the dimensions of microarray data in prostate cancer, leukemia and gastric cancer in order to eliminate redundant attributes and improve the accuracy of the classification process. The reduction process is carried out using Minimum Redundancy Maximum Relevance in the FCQ and FCD equations by forming the k best features values 10, 20, 30, 40, 50, 60, 70, 80, 90 and 100. Meanwhile the classification process is carried out using Random Forest by forming a decision tree of 100 n_estimators. After doing all the processes, the best accuracy for prostate cancer data classification is with an FCQ of 100% at k = 10, without reduction of 95% and the lowest accuracy for FCD is 52% at k = 90. The best accuracy for leukemia data classification is with an FCQ of 93% at k = 20, without reduction 64% and the lowest accuracy is FCD of 57% at k = 80. Finally, the best accuracy for gastric cancer data classification is FCQ and FCD of 100% for all k and the lowest accuracy is without reduction by 83%.