Public transport delays remain a persistent challenge, affecting public transport efficiency and commuter satisfaction. This review explores the potential of smartphone sensor data, combined with machine learning (ML), to enhance delay prediction and mitigation. By systematically reviewing literature on sensor data types, quality, and application, this research identifies optimal data sources for ML-driven transport models. Synthesizing findings from 29 peer-reviewed studies, the review identifies dominant sensor modalities, explores variations in data accuracy across devices, and evaluates ML algorithms applied to delay prediction and road anomaly detection. Findings highlight correlations between sensory data characteristics and predictive accuracy while addressing challenges such as privacy concerns and data-sharing agreements. Despite limitations due to variability in methodologies and technological advancements, this review establishes a foundation for future research into AI-enhanced intelligent transportation systems. Directions for future research include experimental deployments of real-time ML systems, calibration protocols across smartphone hardware, and longitudinal studies assessing data reliability over time.