Training and Testing Healthcare Models using Machine Learning Graphical User Interface Development Environment (ML-GUIDE) Software

Background Machine learning (ML) is transforming healthcare research, but developing models can be complex and time-consuming. We created ML-GUIDE, a no-code tool with a graphical interface, to make ML model building faster and more accessible. Methods We developed the ML-GUIDE software using Python language and the Model-View-Controller architecture. ML-GUIDE provides 40 regression and 25 classification algorithms. We used publicly available Pima Indians Diabetes dataset to assess efficacy of the software. As per literature on ML models for Pima dataset, we developed, trained, and compared results from five diabetes prediction models: Adaboost, Decision Tree, Logistic Regression, Random Forest Classifier, and Support Vector Classifier for this study. We used metrics such as accuracy, average precision, and F1 score to evaluate the performances of ML models. Results We found that the model developed using ML-GUIDE software performed similar to published models. The variability (minimum to maximum) in accuracy of models in literature compared to models developed in this study ranges from 0.15% to 7.54%. The study also revealed that logistic regression with accuracy (78.65%), F1 score (0.849), and average precision (0.776) outperformed other models in predicting diabetes. Conclusion The results demonstrated that ML-GUIDE—a coding free tool matches performance of manual, time-consuming and expert driven algorithms. Therefore, ML-GUIDE tool can be used to rapidly test, develop and deploy ML-based healthcare solutions.