Implementation and Evaluation of Artificial Intelligence -based Models for Disease Prediction in Electronic Health Record (EHR) Systems
Abstract
The incorporation of Artificial Intelligence (AI) in applications related to the Electronic Health Record (EHR) systems has given way to a new innovation in the predictive health analytics. This paper covers the development and the testing of a number of AI computational models within a framework that is described in this study for the use in disease prediction on EHRs. Thus, a controlled selection of possible values for patient’s health was made, and the models of Logistic Regression, Decision Trees, Random Forest, as well as the Neural Networks, were trained using the chosen dataset. To compare the results of each model, four key values, namely accuracy, precision, recall, and F1 Score were employed. In the Neural Networks model apparent retention capability was identified to perform highest with over 92% confidence level on the tested models. The implementation was performed in a simulated EHR environment in order to mimic practice capture requirements of healthcare real-world. The system also incorporates a patient-enabled process enabling the patient to input his/her details and the system produces expected results including confidence level to aid clinical decision. Further comparison with conventional rule-based EHR systems also shown superiority in the forecasting performance which also indicates the uniqueness of the proposed AI incorporated architecture. This paper discusses how AI has the capability of making EHR non-purposeful systems smart diagnostic systems with the potential for personalized healthcare and timely detections. Further studies should address the applied time-based implementation, ensemble with IoT-based medical gadgets, and increased explainability of the model for clinical purposes.