Optimization of backpropagation using nguyen-widrow and stimulus-sampling for breast cancer classification and feature selection

Breast cancer is one of the leading causes of death in the world for women. Early detection and diagnosis can be done to reduce the cancer death rate. MicroRNA is known as a biomarker for breast cancer and with the help of artificial neural network technology, made it possible to perform a classification process for early detection. Backpropagation neural network has good performance in classification, however, still has a drawback related to its long training time. This research is conducted to classify breast cancer (whether a cell is cancer or normal and whether it is before or after metastatic stage) based on microRNA profiles using backpropagation with Nguyen-Widrow and Stimulus-Sampling algorithm optimization. In this paper, three breast cancer datasets are used to compare the classification performances. Furthermore, some alternatives microRNA features set are obtained using feature selection methods and compare the accuracy values. The results show that the combination of Nguyen-Widrow and Stimulus-Sampling algorithm produces the best backpropagation performance based on the accuracy, sensitivity, specificity, and AUC value as well as reducing the running time. The combination of Nguyen-Widrow and Stimulus-Sampling algorithm proved to be able to increase the performance of the method. Nguyen-Widrow algorithm provides weights value initialization that is not too small or too large so that the convergence process can be enhanced. Meanwhile, the use of Stimulus-Sampling improves the performance of backpropagation by strengthening the output unit, which give the smallest error value.