Boundaries tuned support vector machine (BT-SVM) classifier for cancer prediction from gene selection.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Microarray data analysis is crucial for identifying cancer-related genes, but faces challenges with large datasets and conventional data mining techniques due to space and time complexity.
• The dynamic nature of cancer necessitates efficient methods for analyzing vast amounts of gene expression data.

Purpose of the Study:

• To develop an advanced feature extraction and selection method for improved cancer classification using big data approaches.
• To enhance the accuracy and efficiency of cancer detection from DNA microarray data.

Main Methods:

• Implemented Improved Supervised Principal Component Analysis (ISPCA) for feature extraction from gene expression data.
• Utilized a co-variance matrix for gene expression analysis and cancer classification via IPSCA.
• Employed Boundaries Tuned Support Vector Machines (BT-SVM) and a modified Particle Swarm Optimization with a novel wrapper model for further feature selection.

Main Results:

• The proposed ISPCA method demonstrated optimal accuracy, recall, and precision across six benchmark DNA microarray datasets, including leukemia, breast, brain, colon, and lung cancers.
• The system achieved improved performance and reduced training time compared to traditional techniques, both with and without feature selection.

Conclusions:

• The ISPCA-based approach offers a robust and efficient solution for cancer classification from large-scale gene expression data.
• This study highlights the effectiveness of advanced feature selection and extraction techniques in big data analysis for improved cancer diagnostics.