Building Trustworthy Cardiac Models: Cloud-Based Feature Engineering and Software Testing Strategies

Abstract

The development of healthcare management systems (HMS) is led by artificial intelligence and machine learning technology. Heart Disease Prediction (HDP) is a critical aspect of predictive patient care, with the need for precise and reliable prediction models. This article reports an AI/MLOps methodology to HDP, overcoming the defects of conventional methods with data quality, feature engineering, and CI/CD pipeline design considerations. We resolve the problem of inconsistency of data in open-source datasets (e.g., UCI) by a two-phase feature selection. This consists of the hybridization of filter-based (Chi-square, FCBF, Gini Index, ReliefF) and wrapper-based (BFE, EFS, FFS, RFE) approaches for optimal feature set identification. We automate model training, validation, and deployment with reproducibility and scalability as per MLOps best practices. In addition, we stress software quality assurance in the way of systematic testing, such as data validation, performance model testing, and security testing. We test the planned system against the UCI heart disease dataset to demonstrate enhanced prediction accuracy and stability of the model with rigorous experimental validation and deployment scenarios.