Development of a novel hybrid forecasting model for electronic monitoring system planning and application

The electronic monitoring systems that support the surveillance and supervision of suspects, defendants, and convicts through electronic communication methods, while ensuring the protection of victims and society, are a technology increasingly employed in the execution of sentences in advanced countries. Electronic monitoring systems contribute to the effective implementation of probation at the same time as the execution of convicts' sentences in the community. Nowadays, it is observed that crime rates are steadily increasing not only in advanced countries but also in developing nations. In line with the findings, it has been observed that some countries use this system for a more effective execution of the sentence, while some countries use it as an economic alternative to imprisonment. The use and importance of this technology is increasing day by day in Türkiye, which is among the developing countries. These systems play an important role in the execution of sentences and the fight against crime. This thesis study aims to develop a sensitive prediction model for the planning of electronic monitoring systems, an area of current focus. The model is designed to predict the infrastructure requirements of these systems based on historical data of individuals under electronic monitoring. In this context, a hybrid model was created to estimate the number of active offenders in electronic monitoring systems in the short term by using the data obtained from 56.611 people monitored between 2013-2021 by the Electronic Monitoring Branch under Directorate of the Probation Department of the Directorate General For Prisons and Detention Houses of the Ministry of Justice of the Republic of Türkiye. The primary objective of this model is to facilitate efficient planning of the necessary equipment and to ensure the optimal management of electronic monitoring systems in Turkey. Within the scope of the thesis study, the CEEMDAN-Kernel-Meta-ELM hybrid model was proposed as an innovative prediction model and the obtained data was applied using this model. In this thesis evaluate the performance of the proposed model by comparing it with contemporary deep learning methods, traditional methodologies, and other hybrid models. We follow a two-stage approach. In the first stage, we establish the relationships of the implemented models with historical data. In the second stage, we enhance the prediction model's performance by decomposing the data into subcomponents using the CEEMDAN decomposition method. Each subcomponent is subsequently incorporated into the Kernel-based Meta-ELM prediction model. All models are implemented by using the Matlab software environment and present their performance statistically in tables and graphs, using error performance metrics and a Taylor diagram. We engage in a comprehensive comparative discussion of the results to highlight the model's effectiveness and contributions to the field.