A Deep Neural Network Approach for Cancer Types Classification Using Gene Selection

Abstract

Florence Omada Ocheme, Hakeem Adewale Sulaimon and Adamu Abubakar Isah

Cancer classification research is one of the significant areas of exploration in the clinical field. Exact forecasting of various tumor types is an extraordinary challenge and giving an exact forecast will have incredible worth in giving better treatment to the patients. In recent years, many analysis-based investigations have led to the revelation of information on disease subtypes, that has generated high throughput innovations Lately, researchers have attempted to dissect a lot of microarray information for getting significant data that can be utilized in malignancy grouping. To accomplish this objective, one can utilize K-Nearest Neighbor, Neural Networks, Decision Tree, Support Vector a  that would provide approaches needed to break down microarray information towards the choice of best separating quality called biomarker. These machine learning methodologies had the inherent ability to represent the time varying behavior of the underlying biological network that allows for a better representation of spatiotemporal input-output dependencies. Therefore, the exploitation of time series data regarding deep learning has to have become a valuable strategy for deciphering stochastic processes, such as gene expression and classification. Therefore, in this study, another intriguing strategy is introduced to improve the performance of neural networks utilizing deep autoencoder neural networks. This was accomplished through the choice of the first, relevant data, which was being extracted with a Deep Neural Network hidden layer used to train an autoencoder for the classification of the cancer malignancy based on the second stack autoencoder network. The outcome from the proposed experiment was evaluated with the current techniques. Overall, the proposed deep autoencoder accomplished classification accuracy of 99.2% as against the current Modified KNN and SVM which obtained 96.1% and 98.1% respectively.