In recent decades, modeling and forecasting have played an essential role in various applications. Various forecasting models have been developed to improve management, industry, medical sciences, engineering, and data-based decision-making processes. Although these models differ in many respects, they are all based on the assumption that greater generalizability of results is achieved with more accurate models. This means that accuracy is considered the only significant feature of eva‎luating the generalizability of forecasting models. On the other hand, due to the ever-changing conditions of today's world, it is necessary to create a stable and reliable performance to make appropriate decisions. Therefore, the reliability of the models is another factor affecting the generalizability of the model that should be considered in the forecasting models. This thesis has introduced a new forecasting approach based on reliability to solve this gap research. The logic of the proposed approach is based on the fact that the forecasting model can provide more reliable performance to managers and decision-makers with fewer changes accuracy in validation data in different situations. The proposed approach is applied to linear statistical time series models. To eva‎luate the effectiveness of the proposed reliability-based approach, three sets of UCI data have been used, and the results have been compared with time series models based on traditional accuracy. Empirical results show that the proposed approach performs better than the traditional approach in error criteria. It has the same error as the mean squares and the mean absolute error. Also, the statistical test results of significance T show that the progress of the proposed approach is significant compared to time series models based on the traditional approach. Therefore, the proposed approach can be a suitable alternative to the traditional approach of time series models for decision-making. Based on the results, the proposed approach can be an excellent alternative to the traditional approach of time series modeling in real-world problems, especially when more generalizability or reliability is needed.

مهدي خاشعي

علي زينل همداني، صبا صارمي نيا