مهدی خاشعی

The functional logic of classifier models is based on the principle that, to maximize their ability to generalize—an essential factor affecting decision quality in real-world problems—it is crucial to minimize the classification error rate of available historical data. In other words, accuracy is considered the only factor affecting the generalizability of classification methods. However, due to fluctuations in financial variables, stable and reliable forecasts are also necessary for correct and profitable decision-making. Despite the importance of the reliability factor in creating stable and robust results, it has been neglected in the literature on modeling and classification. To address this research gap and enhance decision-making processes in financial applications, a modeling method based on reliability maximization is presented. This paper develops a multilayer perceptron model with the aim of maximizing reliability rather than accuracy. To evaluate the performance of the proposed model, five different financial datasets are selected from the UCI database, and its classification error rate is compared with that of the conventional multilayer perceptron model. The findings show that the reliability factor has a greater impact than the accuracy factor on the generalizability and performance of classification models. The results indicate that the proposed reliability-based multilayer perceptron model demonstrates superior efficiency and performance compared to the conventional multilayer perceptron model and can serve as a viable alternative in financial applications.