Performance Evaluation of Classification Algorithms on Diagnosis of Breast Cancer and Skin Disease

Abstract

Health is so important for human beings. Thanks to the technological developments both in medicine and information technologies, the success percentages of both medical diagnosing and medical treatment systems are increasing day by day. Cancer is the most common causes of death in today’s world and is generally diagnosed at the last stages. Cancer has many types such as breast cancer, skin cancer, leukemia and etc. Diagnosis of cancer at early stages is very important for the success of medical treatments. The aim of this study was to evaluate the classification performances of some popular algorithms on the way to design an efficient computer aided breast and/or skin cancer diagnosing system to support the doctors and patients. For this purpose, same machine learning and deep learning algorithms were applied on immunotherapy dataset and breast cancer Coimbra dataset from UCI machine learning data repository. Feature selection by information gain and reliefF were applied on datasets before classification in order to increase the efficiency of classification processes. Support Vector Machines (SVM), Random Forest (RF), Recurrent Neural Network (RNN) and Convolutional Neural Network (CNN) algorithms were used for classification experiments. Accuracy values are used for performance metric. According to these results, RNN has shown the best performance among the others with 92% on both datasets. This shows that deep learning algorithms especially RNN has great potential to diagnose the cancer from dataset with high success ratios.