Multi-Class Classification of Medical Data Based on Neural Network Pruning and Information-Entropy Measures

Area of Science:

• Medical data analysis
• Machine learning in healthcare
• Computational biology

Background:

• Medical data analysis involves diverse sources like clinical trials and patient-generated health data.
• Machine learning offers powerful tools for classification and clustering but requires suitable architectures for specific tasks.
• Finding optimal machine learning architectures for medical data classification remains a challenge.

Purpose of the Study:

• To propose and evaluate a novel machine learning model for multi-class classification of medical data.
• To investigate the efficacy of information-entropy measures in guiding discriminative pruning within neural networks for medical data processing.
• To apply the proposed model to cancer research using three distinct cancer databases.

Main Methods:

• Developed a two-component machine learning model: a restricted Boltzmann machine and a classifier system.
• Implemented a discriminant pruning method to select salient neurons in the hidden layer, enabling feature selection.
• Utilized information-entropy-inspired dissimilarity measures for post-training neuronal pruning.

Main Results:

• The proposed model demonstrated favorable results after neuronal pruning across three cancer datasets.
• Achieved error rates of 10%-15% for Breast Cancer (compared to 10.68% reported).
• Achieved error rates of 4%-6% for Cervical Cancer (compared to 31% reported) and 31% for Primary Tumour (compared to 20.35% reported).

Conclusions:

• The proposed machine learning model with information-entropy-guided pruning is effective for medical data classification, especially in oncology.
• Discriminative pruning using information-entropy measures enhances neural network performance for cancer data.
• The model shows significant improvements in classification accuracy for cervical and primary tumor datasets.