A Systematic Literature Review of Machine Learning, Deep Learning, and Explainable AI for Predicting Mechanical Performance and Microplastic Release in Plastic Waste-Modified Asphalt

The growing pile of plastic waste has led to the implementation of sustainable engineering solutions, such as putting recycled plastics into asphalt mixtures, referred to as Plastic Waste-Modified Asphalt (PWMA). Despite the fact that in earlier studies, a significant improvement in mechanical performance, particularly stiffness and rutting resistance, has been demonstrated, there has been no emphasis on the environmental consequences related to it, particularly microplastic release (MPs). In line with this, Machine Learning (ML) and Deep Learning (DL) models have been highly used in estimating the performance of pavement with a high level of accuracy, but their non-conventional characteristics make them hard to implement in the engineering profession. It is a systematic literature review that is based on PRISMA 2020 and aims to summarize the most up-to-date state-of-the-art in the crossroads of PWMA, advanced computational modeling, and environmental sustainability. 36 peer-reviewed articles that were published in 2019-2026 were examined. The results indicate that the DL models (particularly Multi-Layer Perceptron (MLP) and Neural Architecture Search-based models) have high predictive accuracy (R2 > 0.95), which is higher than the traditional ML methods. However, 32 of the 36 reviewed studies (88.9%) primarily focused on mechanical performance, whereas only 4 studies (11.1%) addressed microplastic-related environmental issues. In addition, there is an explainability gap, with only 4 of the 36 studies (11.1%) incorporating Explainable Artificial Intelligence (XAI) methods. A suspected trade-off that is of critical importance in this paper is the possibility of a higher content of plastics to result in a higher stiffness but potentially higher rates of microplastic formation due to the brittle nature of materials. To overcome these shortcomings, a new Explainable Deep Learning (XDL) architecture is suggested that allows predicting mechanical performance and microplastic emissions simultaneously with greater interpretability. The specified strategy will be used to develop data-driven, transparent, and sustainable pavement design strategies.