An Efficient PCA-GA-HKSVM-Based Disease Diagnostic Assistant.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Computational Biology

Background:

• Disease diagnosis is hindered by misdiagnosis, delays, and missed diagnoses.
• Machine learning classification models use symptoms to predict disease presence, but performance can be improved.

Purpose of the Study:

• To present an improved machine learning technique for disease diagnosis.
• To enhance classification model performance using feature selection and hybrid kernel methods.

Main Methods:

• Feature selection via Principal Component Analysis (PCA).
• A hybrid kernel-based Support Vector Machine (HKSVM) combining Radial Basis Function (RBF), linear, and polynomial kernels.
• Hyperparameter optimization using a Genetic Algorithm (GA).
• The combined PCA-GA-HKSVM model was evaluated on 7 medical datasets (2 multiclass, 5 binary).

Main Results:

• The PCA-GA-HKSVM demonstrated superior performance compared to single-kernel Support Vector Machines (SVMs).
• Evaluations using accuracy, precision, and recall metrics confirmed the model's effectiveness.
• Combining local (RBF) and global (linear, polynomial) kernels improved model performance by better distinguishing data points at different ranges.

Conclusions:

• The proposed PCA-GA-HKSVM technique offers a significant advancement in disease diagnostic accuracy.
• Hybrid kernel approaches in machine learning can effectively address complex medical diagnostic challenges.
• This method provides a more robust and reliable tool for medical diagnosis systems.