Evolutionary and Neural Computing Based Decision Support System for Disease Diagnosis from Clinical Data Sets in...

Area of Science:

• Computational intelligence
• Machine learning
• Clinical decision support

Background:

• Large clinical databases offer insights but contain irrelevant data, hindering accurate disease diagnosis.
• Existing machine learning approaches can be confused by noisy or incomplete data, degrading prediction outcomes.

Purpose of the Study:

• To develop an intelligent approach for disease diagnosis using an optimal subset of attributes.
• To enhance clinical decision-making by improving the accuracy and efficiency of diagnostic procedures.

Main Methods:

• An Application Specific Intelligent Computing (ASIC) system was developed, featuring a two-phase approach.
• Phase 1: Rough set-based genetic algorithm for data pre-processing, handling outliers, noise, and missing values to select optimal attribute reducts.
• Phase 2: Back propagation neural network for training and testing classifiers on the selected attributes.

Main Results:

• The ASIC system achieved high accuracy rates on benchmark datasets: 95.33% for breast cancer, 97.61% for fertility diagnosis, and 93.04% for heart disease.
• Performance evaluation demonstrated that the proposed ASIC system outperformed existing classifiers.

Conclusions:

• The proposed ASIC system effectively preprocesses clinical data, identifies optimal attributes, and enhances diagnostic accuracy.
• This intelligent computing approach shows significant promise for improving clinical decision support systems.