Analysis of health data with heuristic hybrid learning methods

The analysis of health data holds critical importance in the diagnosis and prediction of diseases. With the increasing volume of data in today's world, there is a clear demand from researchers and healthcare professionals for accurately designed diagnostic systems using machine learning methods. However, machine learning methods, being designed based on balanced datasets and complete information, often lead to erroneous results in healthcare datasets due to their inherently imbalanced and incomplete nature. In this thesis, a hybrid preprocessing method has been developed to tackle class imbalance and missing value problems. This method includes the GA-MICE approach, which utilizes Genetic Algorithm (GA) heuristics and the Multiple Imputation by Chained Equations (MICE) method for completing missing values. Additionally, it incorporates the GASMOTEPSO_ENN method, which combines GA and Particle Swarm Optimization (PSO) heuristics with Synthetic Minority Over-Sampling Technique (SMOTE) and Edited Nearest Neighbors (ENN) undersampling technique for balancing imbalanced class distributions. The effectiveness of the proposed method was tested on publicly available datasets for significant health issues such as diabetes, stroke, and kidney disease using six different machine learning classification methods. The findings revealed that the proposed method achieved accuracy, precision, recall, F1-score, and Area Under the Curve (AUC) values ranging from 93% to 100% across the three datasets. This method effectively addressed class imbalance and missing value issues, yielding higher and more reliable results compared to similar methods in the literature.