An Integrated Machine Learning Framework for Early Detection of Dialysis Complications Using SMOTE, Ant Colony Optimization, and Ensemble Classifiers

Abstract:

Dialysis related problems have been a prominent obstacle for end stage renal disease (ESRD) patients. In this paper, an end to end machine learning model was propo sed to early predict dialysis complications by utilizing a simulated but clinically valid data. It includes data pre processing, class re
sampling of Synthetic Minority Over sampling Technique (SMOTE) to handle the data imbalance, and Ant Colony Optimizati on (ACO) algorithm was applied for the feature selection, selecting Creatinine, Potassium and Hemoglobin as the features of high importance. Support Vector Machine (SVM),
Random Forest (RF) and Artificial Neural Network (ANN) classifiers were trained, eval uated on the criteria of accuracy, precision, recall, F1 score, ROC AUC and Cohen's Kappa. RF classifier reached higher performances: 99.70% accuracy, 100% precision, 99.25% recall, 99.62% F1 score, and
0.9999 ROC AUC. These findings prove that effective p rediction of dialysis complications can be achieved by using the combination of SMOTE, ACO based feature selection and ensemble learning methods. The utilization of real world clinical datasets to evaluate and demonstrate the proposed method will be necess ary to make sure the feasibility and generalizability before deployment in clinic. This research has made contribution to est ablish the intelligence decision supporting systems in nephrology field.

Keyword: Dialysis, End Stage Renal Disease (ESRD), Ant Colony Optimization (ACO), Synthetic Minority Over Sampling Technique (SMOTE), Support Vector Machine (SVM), Artificial Neural Network (ANN), Random Forest (RF) and Cohen s Kappa.