This study proposes a novel approach to driver drowsiness detection using the Video Vision Transformer (ViViT) model, which captures both spatial and temporal dynamics simultaneously to analyze eye conditions and head movements. The National Tsing Hua University Driver Drowsiness Detection (NTHU-DDD) dataset, which consists of 36,000 annotated video clips, was utilized for both training and evaluation. The ViViT model is compared to traditional Convolutional Neural Network (CNN) and Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) models, demonstrating superior performance with 96.2% accuracy and 95.9% F1-Score, while maintaining a 28.9 ms/frame inference time suitable for real-time deployment. The ablation study indicates that integrating spatial and temporal attention yields a notable improvement in model accuracy. Furthermore, positional encoding proves essential in preserving spatial coherence within video-based inputs. The model’s resilience was tested across a range of challenging conditions including low-light settings, partial occlusions, and drastic head movements and it consistently maintained reliable performance. With a compact footprint of just 89 MB, the ViViT model has been fine-tuned for deployment on embedded platforms such as the Jetson Nano, making it well-suited for edge AI applications. These findings highlight ViViT’s promise as a practical and high-performing solution for real-time driver drowsiness detection in real-world scenarios.

Pavement distress is a critical factor in road maintenance planning, directly influencing transportation safety, serviceability, and infrastructure costs. While traditional mechanistic and statistical models provide limited accuracy, they often fail to capture the nonlinear and multi-factorial nature of pavement deterioration. This study addresses this gap by proposing an integrated machine learning (ML) framework that incorporates real-time traffic and climatic variables for predicting pavement roughness. The framework draws on multiple open-source datasets, Long-Term Pavement Performance (LTPP), Federal Highway Administration (FHWA) traffic volumes, and National Oceanic and Atmospheric Administration (NOAA) climate records, to construct a multidimensional feature space. Four predictive algorithms were benchmarked: Random Forest (RF), XGBoost (XGB), Support Vector Machine (SVM), and Multi-layer Perceptron (MLP). Ensemble-based models achieved superior predictive accuracy, with Random Forest attaining R$^2 \approx$ 0.89 and Root Mean Square Error (RMSE) $\approx$ 0.61, outperforming traditional regression baselines. The findings highlight that ensemble learning can more effectively capture non-linear dependencies between structural, traffic, and climatic factors than alternative approaches. Beyond technical performance, the study illustrates the potential of integrating continuously updated environmental and traffic data into pavement management systems, offering a pathway to more cost-efficient, reliable, and sustainable maintenance planning.

Aceh Province is a critical case for freight and infrastructure studies due to its geographic isolation, post-disaster recovery context, and heavy dependence on roads for over 95% of commodity transport. Despite its rich agricultural output, limited multimodal infrastructure hampers efficient distribution. This study aims to (1) analyze the effect of road network connectivity on commodity transportation and regional development, and (2) develop a forecasting model to predict future commodity transportation needs in Aceh Province. The Structural Equation Modeling (SEM) was applied to analyze the relationships among Road Network Connectivity (RNC), Freight Transport (FT), and Regional Development (RD), using data from 400 respondents across 23 districts. The SEM results show all latent variables are interconnected. FT plays a strong mediating role, linking connectivity improvements to development benefits. The study also develops forecasting models for commodity generation and attraction based on population, expressed as $Y$ = 2.209 $X_1$ and $Y$ = 2.807 $X_1$. These models highlight population as a reliable predictor of freight demand and can be generalized to other regions with similar geographic and infrastructure constraints. This research introduces a novel SEM-based framework for freight analysis in Indonesia and offers policy insights for integrating road infrastructure planning with regional development strategies.

Maritime transportation safety is a strategic priority in Indonesia’s national logistics system as an archipelagic country. However, high shipping intensity and limited inspection resources pose serious challenges to the effectiveness of ship inspections. Ship Safety Inspectors (PPKK) play a crucial role in ensuring the seaworthiness of ships through inspections of nautical, technical, and communication (radio) aspects; however, their functional contributions have rarely been systematically examined. This study employs a quantitative approach using survey methods and Structural Equation Modeling–Partial Least Squares (SEM-PLS) analysis techniques. A sample of 64 CSOs from three strategic ports (Tanjung Priok, Soekarno-Hatta, Makassar, and Sorong-Pelra) was analyzed to examine the relationship between inspection functions, workload, and maritime safety. The results indicate that technical, nautical, and communication functions significantly influence workload, while technical functions and workload have a direct impact on maritime safety. Work volume acts as a mediating variable in the model. The implications of these findings emphasize the importance of enhancing the competence of PPKK and developing risk-based inspection policies. This study provides empirical contributions to the reform of the maritime safety supervision system and opens opportunities for cross-national validation to strengthen the generalizability of the model.

Efficient light rail transit (LRT) systems are crucial for sustainable urban mobility; however, unforeseen departure delays continue to be a major hurdle, undermining operational reliability and passenger satisfaction. This study establishes a data-driven framework for forecasting departure delays by combining static GTFS schedules with real-time GTFS operational data from the Canberra LRT system. The dataset included 15,538 records with 42 attributes, spanning from 28 August 2020 to 13 August 2022. A stringent preprocessing pipeline was implemented, encompassing temporal feature engineering and feature selection based on mutual information. The Random Forest regressor with feature engineering and selection (RFR-FEFS) attained the highest predictive performance on the test set ($R^2$ = 0.94, MAE = 2.93, MSE = 34.32). The high accuracy indicates the model’s efficacy, yet it necessitates careful evaluation of potential overfitting and its generalizability beyond the examined system. Ablation experiments were performed to assess the impact of various feature groups by omitting temporal, spatial, or operational attributes. The findings indicate that the exclusion of temporal features decreased $R^2$ to 0.90, the exclusion of spatial features reduced it to 0.93, and the exclusion of operational features resulted in the most significant decline to 0.23. These findings affirm that all three feature categories contribute distinctly and synergistically to model performance. This research illustrates the capability of integrating diverse GTFS data with sophisticated machine learning techniques to attain precise LRT delay forecasts. Nevertheless, the framework was validated solely on one system and time frame; future research should investigate its transferability to other cities and integrate supplementary contextual data, including meteorological conditions and incident reports, to improve robustness and practical applicability.