Community Perceptions and Machine Learning Analysis of Environmental Stressors on Active Transportation and Micromobility in Delaware

Environmental stressors including coastal inundation, rising temperatures, flooding, and urbanization increasingly threaten non-motorized transportation facilities (NMTFs) and micromobility systems. A companion study used GIS-based geospatial modeling to quantify the physical exposure of trails and bike routes across Delaware under multiple environmental scenarios. The present study complements that geospatial assessment with a human-centered analysis, combining a 200-respondent public survey with supervised machine learning to characterize community perceptions of environmental disruption and to identify the factors that predict two key outcomes: perceived safety of active transportation and perceived availability of micromobility. Survey responses were transformed into a structured feature set spanning environmental exposure frequency, urbanization pressure, demographic characteristics, and travel-mode dependence. Random Forest and Decision Tree classifiers were trained and evaluated using five-fold stratified cross-validation. The Random Forest model predicted perceived safety with 61.3% cross-validated accuracy and micromobility availability with 65.5% accuracy, outperforming the simpler Decision Tree in both tasks. Feature importance analysis identified cumulative environmental impact and flooding frequency as the strongest predictors of perceived safety, while income level and environmental exposure were the leading predictors of micromobility availability. Descriptive results show that 70% of respondents regard micromobility infrastructure as insufficient and that over 90% of respondents prioritize physical infrastructure investment over financial or educational interventions. These findings, interpreted alongside the geospatial vulnerability results, indicate that environmentally exposed communities in Delaware experience a compounding disadvantage of physical and perceptual mobility risk, underscoring the value of coupling AI-driven survey analytics with geospatial exposure modeling in transportation planning. Correction: the phrase “outperforming the simpler Decision Tree in both tasks” above should be read as applying to the safety-prediction task; a Decision Tree comparison for the micromobility-availability task was not performed in this study, and this omission, along with a possible target-leakage issue affecting the reported Model 2 accuracy, is discussed as a limitation in Section 5.3.