Enhancing air quality classification using a novel discrete learning-based multilayer perceptron model (DMLP)

Effective utilization of data analysis techniques is paramount in addressing the complex challenges presented by environmental issues. These methodologies empower researchers and practitioners to derive meaningful insights from intricate datasets encompassing air quality, biodiversity, climate change, and other pivotal environmental factors. Through the deployment of robust classification models, such as intelligent classifiers, researchers can accurately classify and predict environmental phenomena. This capability holds significant implications for guiding policy decisions, mitigating environmental risks, and devising sustainable solutions to protect our natural resources and ecosystems. Thus, classification models not only deepen our comprehension of environmental dynamics but also empower proactive measures towards achieving environmental sustainability and resilience amidst global challenges. Intelligent classifiers, distinguished by their exceptional capabilities, have demonstrated superior performance compared to other classification models. However, in all developed intelligent classifiers a similar cost/loss function is implemented in the learning processes, which is continuous and works based on the distance between actual and fitted values. Whereas the nature of the classification is discrete. As a result, in this study, a novel cost/loss function is proposed that in contrast to its conventional version is discrete and works based on the direction. In order to explain the process of the suggested methodology, the feed-forward multilayer perceptrons that are among the most famous intelligent classifiers is considered. In this paper, in order to determine the superiority of the proposed model in the domain of environment, it is implemented on some benchmark data sets which is related to air quality. Numerical results indicate that the performance of the proposed model is better than the conventional multilayer perceptrons in whole benchmark data sets. In addition, numerical results clarify that the developed discrete learning-based multilayer perceptron classifier can averagely gain an 87.68% classification rate, which points to more than 9% improvement over its conventional version.