Bayesian LSTM-Based Missing Data Estimation and Flexural Strength Assessment for Determination of Novel Smart Mobility Pavement Materials: Mg(OH)2 Added Plastic Composites

The construction field needs to reduce carbon emissions; therefore, many methods are being attempted applying to materials research. Concrete and asphalt are the representative materials for the pavement system. However, a lack of aggregates and certain limits in binder replacement are the main obstacles to achieving a reduction in carbon emissions. This study aims to address the lack of sustainable alternatives in pavement materials by investigating recycled plastic composites with Mg(OH)₂, thereby filling a research gap in low-carbon, mechanically viable solutions for smart mobility infrastructure. Three types of recycled plastics-polypropylene (PP), polyethylene terephthalate (PET), and OTHER (OTH) resins-were combined with Mg(OH)₂ to produce nine specimen configurations. The mechanical behavior and flexural performance were evaluated through displacement-flexural stress curve tests, while missing experimental data were reconstructed using a Bayesian long short-term memory (BLSTM) machine learning approach. The BLSTM model achieved an average R² of 0.8018 in testing and 0.7618 in validation, confirming reliable prediction capability even with a small dataset. All composites demonstrated a minimum flexural strength of 30 MPa, with PP-based specimens reaching approximately 40 MPa, confirming their suitability for pavement applications. These results highlight the flexural performance of each composite type, with PP emerging as the most promising candidate.