Analysis of the Impact of Dimensionality Reduction on Deep Learning Performance for the Classification of Microarray and RNA-Seq Gene Expression Data

One of the most popular and widely used methods in various fields, not only in information technology, including bioinformatics. With the increasing complexity of data and problems faced, ML offers solutions for analyzing and interpreting very large and dense data. This study aims to analyze gene expression, which has a crucial role in disease diagnosis, especially cancer, requiring techniques capable of identifying patterns and relationships in complex genetic data. Several techniques that play a role in improving the performance of deep learning models, including PCA reduction techniques, TruncatedSVD, PaCMAP, and TorchDR, successfully simplified data with high dimensionality without eliminating important information structures, thereby speeding up the training process and improving prediction accuracy. DNN was able to achieve a perfect score with accuracy and F1-Score 1 and remained superior to TorchDR with accuracy using test data of 0.9990, PaCMAP 0.979, and TruncatedSVD 0.988. DNN is able to obtain a perfect score with accuracy and F1-Score 1, and remains superior to TorchDR with accuracy using test data of 0.9990, PaCMAP 0.979 and TruncatedSVD 0.988. The dimensionality reduction techniques of PaCMAP and TorchDR provide a more stable data representation, by producing a more separated data distribution and supporting high performance in DNN and MLP models, which makes it the right choice in the dimensionality reduction method of gene expression data.