Imbalanced datasets

So far, distress prediction models have been based on balanced, such sampling is not consistent with the reality of the statistical community of companies. If the data are balanced, the bias in sample selection may lead to an underestimation of typeI error and an overestimation of the typeII error of models. Although imbalanced data-based models are compatible with reality, they have a higher typeI error compared to balanced data-based models. The cost of typeI error is more important to Beneficiaries than the cost of typeII error. In this study, for reducing typeI error of imbalanced data-based models, random and clustering-based undersampling were used. Tested data included 760 companies since 2007-2007 with 4 different degrees and the results of the H1 to H3 test represented them. In all cases of the typeI error, typeII error of balanced data-based models were lower and more, respectively, compared to imbalanced data-based models; also, in most cases, the geometric mean of balanced data-based models was higher compared to imbalanced data-based models, respectively. The results of testing H4 to H6 show that in most cases, typeI error, typeII error and the geometric mean criterion of models based on modified imbalanced data were less, more, and more, respectiively compared to the models based on imbalanced data, in other words, applying Undersampling methods on imbalanced training data led to a decrease in typeI error and an increase in typeII error and geometric mean criteria. As a result using models based on modified imbalanced data is suggested to Beneficiaries

Facial emotion recognition has recently attracted considerable interest due to its wide range of applications. It plays a crucial role in supporting individuals with autism spectrum disorders and improving interactions between humans and computers. The ability to execute these applications in real-time is essential. The architecture of the model and the computational resources available are the key determinants of inference time. Consequently, the development of a real-time solution requires a concentrated effort on these elements. In this paper, we present a scalable approach that utilizes EfficientNetV2, chosen for its operational efficiency. Our methodology involves resolution scaling based on a polynomial equation, which ensures real-time performance across various computational resources and model configurations. This scalable technique employs a polynomial equation to identify the optimal resolution for designated inference times, specifically adapted to our hardware and model specifications. By implementing the polynomial equation for resolution scaling, we created two variants of EfficientNetV2. Our findings from the KDEF dataset indicate that the proposed EfficientNetV2 can accurately classify images in real time on our hardware.