Comparative Analysis of Classification Methods with PCA and LDA for Diabetes

Area of Science:

• Medical Informatics
• Computational Biology
• Machine Learning in Healthcare

Background:

• Diabetes is a leading chronic disease influenced by lifestyle and genetics.
• Early diagnosis is vital for effective management and treatment.
• Developing efficient diagnostic systems is a priority in modern healthcare.

Purpose of the Study:

• To develop an indigenous and efficient diagnostic technique for early diabetes detection.
• To evaluate the performance of various classification methods with and without feature reduction.
• To identify the optimal approach for accurate diabetes prediction.

Main Methods:

• Utilized the Pima Indian Diabetes Dataset (PIDD) and a Localized Diabetes Dataset (LDD).
• Applied classification algorithms: Adaboost, Classification via Regression (CVR), Radial Basis Function Network (RBFN), K-Nearest Neighbor (KNN).
• Implemented feature reduction techniques: Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) prior to classification.

Main Results:

• Compared classification performance with and without PCA and LDA.
• PCA combined with CVR (PCA_CVR) demonstrated maximum performance on both datasets.
• Feature reduction techniques decreased computation time and improved Receiver Operating Characteristic (ROC) curve accuracy.

Conclusions:

• PCA and LDA effectively remove insignificant features, reducing costs and computation time.
• Feature reduction enhances diagnostic accuracy and ROC performance.
• The developed methodology is adaptable for diagnosing other medical conditions.