A Comparative Study of Hybrid Dimension Reduction Techniques to Enhance the Classification of High-Dimensional Microarray Data

This study explores the challenges posed by high-dimensional data in the context of microarray gene expression data, emphasizing the importance of dimension reduction techniques for informative gene selection. The research assesses various hybrid dimension reduction methods involving feature extraction and selection. Using cancer datasets, the study employs principal component analysis (PCA), linear discriminant analysis (LDA), and singular value decomposition (SVD) as dimension reduction techniques. Feature selection methods such as Information Gain (IG) and Lasso regularization are applied to enhance classification performance. Machine learning algorithms, including Support Vector Machines (SVM), Decision Trees, Random Forest, and Naïve Bayes, are employed for classification after feature selection. The evaluation relies on the confusion matrix, measuring accuracy. The finding reveals that SVM excels in binary data classification, while Random Forest outperforms in multiclass data classification. SVM consistently records the highest accuracy across all hybrid techniques, reaching 91.95% accuracy in binary data with LDA. In multiclass data, the SVD-IG-Lasso approach combined with Random Forest achieves a remarkable 100% accuracy. These findings demonstrate the success of the three-tier hybrid approach in effectively handling high-dimensional data challenges. In conclusion, this research offers valuable insights into the classification of high-dimensional data, showcasing SVM's strength in binary data and Random Forest's superiority in multiclass data classification by demonstrating the efficacy of hybrid dimension reduction techniques.