چكيده به لاتين
Service composition is a key method for delivering integrated and patient-centered health services. However, current methods face limitations in creating adaptive and patient-centered solutions due to challenges in real-time data management and the dynamic conditions of service providers. To address this gap, this study proposes a two-stage adaptive and patient-centered service composition model based on the predictive process monitoring method (PPM-APCSC). Utilizing the predictive process monitoring (PPM) approach, the PPM-APCSC model leverages real-time data from patient healthcare systems to dynamically predict subsequent sub-tasks in the care pathway and allocate resources based on Quality of Service (QoS) criteria. The development of this model began with an exploration of different PPM approaches, namely Classification-Based PPM (CB-PPM) and Regression-Based PPM (RB-PPM), to gain a comprehensive understanding of their applicability in service composition. One objective of service composition is to optimally select resources for sub-tasks; however, existing PPM methods often overlook resource status. To address this limitation, this study developed a resource-aware predictive process monitoring approach (RA-PPM) to improve prediction performance and ensure seamless integration into the proposed model. Ultimately, the PPM-APCSC model was developed to facilitate effective service composition, and its advantages and limitations were thoroughly analyzed.
The Design of Experiments (DoE) method was used to analyze CB-PPM and RB-PPM approaches. The performance of these two approaches was evaluated by running 136 models on 10 real-world datasets. The results, examined from the perspectives of accuracy and computation time, indicated that CB-PPM was the superior approach in 90% of cases. Moreover, to evaluate the performance of the RA-PPM model, 156 models with different configurations were tested on 8 real-world datasets. The experimental results demonstrated that incorporating resource awareness in the PPM method enhances prediction accuracy, albeit at the cost of increased online execution time. Finally, the CRISP-DM method was employed to develop the PPM-APCSC model. The PPM-APCSC model leverages event log data to train two PPM models: one for predicting the next sub-task in the service provision pathway, and another for predicting the process completion time for service composition. The evaluation of the results from experiments conducted on three real-world healthcare datasets demonstrated that this model has a high success rate, achieving over 93% feasibility in creating a chain of activities and more than 81% accuracy in predicting the next step of the process. Furthermore, the model effectively predicted process completion time, with a Mean Absolute Error (MAE) of less than 104 minutes. In addition, the PPM-APCSC model generated service compositions that were 87% similar to the best previous service compositions. The efficiency of the PPM-APCSC model was evaluated within an integrated medical diagnostics laboratory network. The results demonstrated that the model could predict sub-tasks with more than 82% precision, and the feasibility of the chain of these predicted sub-tasks was over 97%. Furthermore, the model accurately predicted the process completion time of the samples in the case study, achieving a MAE of less than 7.8 minutes. Based on this, the model provided service compositions that were 92% similar to the best historical service compositions.
