Optimizing Cancer Detection through Gene Expression Data: A Comparative Study of Dimensionality Reduction and Deep Learning Techniques

The rapid evolution of gene expression analysis has brought new opportunities for early and accurate cancer detection, yet the high dimensionality of such data remains a significant challenge. This study explores the optimization of cancer detection through gene expression data by employing a range of dimensionality reduction techniques and deep learning models. By focusing on breast, brain, colon, and other cancer types, this research compares various methods for minimizing data complexity while maintaining or enhancing classification accuracy. Three key approaches are investigated: a direct classification using Support Vector Classifier (SVC), and two advanced methodologies that integrate SVC with novel optimization algorithms, including the Chimp Optimization and its enhanced variant. The study demonstrates that reducing the data's dimensionality through intelligent feature selection significantly improves the performance of machine learning models, resulting in faster, more accurate cancer classification. The integration of deep learning architectures, such as the Depth-wise Separable Convolutional Neural Network (DSCNN), further refines the classification process by effectively handling the selected gene subsets. The findings highlight the critical role of adaptive optimization and deep learning in managing high-dimensional data, enhancing classification speed, and reducing error rates. This comparative analysis underscores the potential of these combined strategies to revolutionize early cancer detection, offering a robust framework that can be adapted to diverse datasets and cancer types, ultimately paving the way for more personalized and precise patient care.