Identifying cancer risks using spectral subset feature selection based on multi-layer perception neural network for...

Area of Science:

• Biomedical research
• Computational biology
• Machine learning in healthcare

Background:

• Cancer poses a significant global health challenge.
• Accurate prediction of cancer risk is crucial for early detection and intervention.
• Feature selection and classification are key challenges in cancer risk analysis.

Purpose of the Study:

• To propose a novel Subset Clustering-Based Feature Selection using a Multi-Layer Perception Neural Network (SCFS-MLPNN) for cancer risk prediction.
• To enhance classification accuracy for early cancer detection.
• To identify and exploit relational features for improved risk analysis.

Main Methods:

• Pre-processing using Intensive Mutual Disease Influence Rate (IMDIR) and Successive Disease Pattern Stimulus Rate (SDPSR) to identify relational features and patterns.
• Feature selection and clustering using Inter-Class Sub-Space Clustering (ICSSC) and spectral subset feature selection (SSFS).
• Classification using a Multi-Layer Perception Neural Network (MLPNN) trained on selected subset features.

Main Results:

• The proposed SCFS-MLPNN method achieved a risk analysis accuracy of 91.8%.
• An F-measure of 91.3% was obtained, indicating high precision and recall.
• The method demonstrated superior classification accuracy compared to previous approaches.

Conclusions:

• The SCFS-MLPNN effectively exploits subset features through relational feature clustering for improved cancer risk classification.
• The proposed method shows significant promise for early cancer detection and diagnosis.
• The high accuracy achieved supports its recommendation for clinical application in early cancer risk assessment.