Genetic Algorithm-Based Feature Selection and Optimization of Backpropagation Neural Network Parameters for Classification of Breast Cancer Using MicroRNA Profiles

Breast cancer is one of the most common types of cancer found in women. Breast cancer mortality increases every year because it has not found an appropriate early detection method. MicroRNA can be used as a potential biomarker, because the profile of the microRNA feature in breast cancer will decrease or increase the value of expression compared to normal conditions. But because of the thousands of types of microRNA that make up breast cancer, a lot of money is needed to detect it entirely. Backpropagation Artificial Neural Network Method has good performance in generalization, so it is suitable to be used as a method for classification with many features. The classification results from the neural network model will be more accurate if the parameters used can be optimized precisely. Genetic algorithms can be used to optimize backpropagation neural network parameters as well as feature selection, because of its global search characteristics. This study aims to compare the performance of backpropagation artificial neural networks optimized parameters as well as feature selection using genetic algorithms (GABPNN_ FS) with backpropagation artificial neural networks optimized using genetic algorithms without feature selection (GABPNN). The results showed that the GABPNN had better results with an error value of 0.016115. But GABPNN_ FS has a faster average process duration of 53.2689 seconds. The best individual chromosome translation results on GABPNN_ FS for breast cancer classification based on microRNA profile are random state = 6098, learning rate = 0.7, number of neuron hidden = 6, and selected features = 707 features that produce accuracy, sensitivity, and specificity ie 97.50 %, 99.00% and 96.00%.