Local Entropy Based Remora Optimization and Sparse Autoencoders for Cancer Diagnosis Through Microarray Gene Expression Analysis

Abstract

Gene expression analysis can be used as a tool to detect cancer and the type of cancer in its early stages. However, the computational complexity for gene expression analysis is quite high due to the large amount of numerical data. Hence, dimensionality reduction techniques are employed to reduce the computational complexity, which in turns reduces the accuracy of cancer diagnosis. This research work focuses on the development of computerized tools with reduced computational complexity and increased accuracy to identify cancer cells in five different regions, namely breast, liver, lung, prostate, and renal. Sparse autoencoder is implemented to reduce feature dimensions, and Remora Optimization is used to increase the performance of cancer diagnosis system through transformation of features. The local entropy-based fitness function is utilized to evaluate the appropriateness of population in Remora Optimization algorithm, and it is considered as the important reason for performance improvement in the proposed cancer diagnosis system. 1027 patients are taken into consideration in this analysis of the microarray datasets, which were gathered from the Curated Microarray Database. Six distinct metrics are used to examine the classification performance, and four different supervised classifiers are investigated. When the suggested sparse autoencoder and local entropy-based Remora Optimization are applied, the average balanced accuracy of the examined classifiers is 93.4%. The average dimensions are decreased from 36045 to 80 dimensions using the suggested method. In the absence of the suggested methodology, the vanilla classifiers yield an average balanced accuracy of 82.7%.

Keywords

• Cancer
• entropy
• sparse autoencoder
• remora optimization
• dimensionality reduction
• deep learning