التحليلات التنبؤية القابلة للتفسير في مجال الرعاية الصحية استناداً إلى تقنيات تنقيب البيانات الزمنية

التحليلات التنبؤية القابلة للتفسير في مجال الرعاية الصحية استناداً إلى تقنيات تنقيب البيانات الزمنية

الباحث. احمد غني داود
جامعة ديالى / كلية التربية للعلوم الانسانية

:Abstract

The convergence of temporal data mining and predictive analytics has heralded the beginning of the era of precision medicine, providing the ability to catch diseases early, personalize treatments, and control epidemic outbreaks more proactively. This paper describes a robust framework that integrates temporal data mining techniques of time-series clustering, sequential pattern mining, and anomaly detection with interpretable predictive models such as decision trees, gradient boosting, and attention-based recurrent neural networks (RNNs). The framework emphasizes interpretability and utilizes SHAP (SHapley Additive exPlanations) values, attention-based mechanisms, and purpose-built visual representations to render predictions understandable and actionable clinically. We demonstrate the feasibility of the framework in three case studies chronic disease management (diabetes), prediction of emergency department (ED) hospital return visits, and epidemic outbreaks, while leveraging publicly available datasets (e.g. MIMIC-III, CMS Hospital Readmission, WHO COVID-19). The proposed framework achieved high performance across all three case studies (e.g. 90% accuracy for epidemic outbreaks, 0.80 area under curve for hospital return visits) and useful insights into predicting healthcare behaviors. Eight tables present methodological and dataset characteristics, justification of feature importance, and measures of performance. Nine graphs presented in distinct color palettes demonstrate temporal trend patterns, layers of model comparisons, and visualization interpretation insights. This framework presents a scalable and transparent approach for health analytics with the aim of increasing the accessibility of complex modelling techniques to promote clinician trust in predictive algorithms.

Keywords: temporal data mining, predictive analytics, explainable AI, healthcare, time-series clustering, sequential pattern mining, recurrent neural networks, SHAP values, attention mechanisms, chronic disease management, hospital readmission, epidemic forecasting