Juan Pineda-Jaramillo

Efficient railway track maintenance is critical for safety, operational reliability, and cost-effective asset management. While traditional inspection methods are well-established and comply with safety regulations, they face challenges in providing continuous, real-time monitoring capabilities that could enhance preventive maintenance strategies. This study introduces a hybrid machine learning framework for real-time anomaly detection in railway tracks, leveraging unsupervised clustering and spectral analysis to improve defect identification. The proposed approach integrates Principal Component Analysis (PCA) and MiniBatch K-Means clustering with a novel distance-from-mean spectral analysis technique to detect deviations in accelerometer signals from in-service locomotives. Vertical and lateral axle-box accelerations undergo Short-Term Fourier Transform (STFT) processing, generating normalized spectrograms that ensure signal consistency. Clustering identifies major anomalies based on spectral pattern shifts, while the distance-based method enhances sensitivity to subtle defects. Experimental results demonstrate that the hybrid model effectively detects 1.5–3.5% of track segments as high-risk across different railway routes. The combination of clustering and spectral deviation analysis enhances anomaly detection sensitivity, improving the identification of track defects. The framework was validated against ground truth data from Metro Valencia's track inspection records on Lines L3 and L9. Results revealed global recall scores ranging from 0.67 to 0.83, and weighted recall scores up to 0.97, depending on the method and track line. Clustering-based detection showed strong performance in identifying track width and lateral deviation issues on L3 (up to 1.00 recall), while the distance-based method excelled in detecting twist-related anomalies on L9. These findings confirm the complementary strengths of both methods in practical railway scenarios. The proposed framework complements existing inspection systems and can support enhanced maintenance planning and early defect detection. Future work will focus on expanding validation campaigns, integrating additional sensor modalities, and refining detection thresholds for broader deployment in transport operations.

This study presents a conceptual framework proposal for integrating urban freight initiatives (UFIs), or city logistics initiatives, into urban planning and urban management (UPUM) land use and infrastructure systems. As a novel approach, this framework integrates three components: Firstly, a conceptual basis on three economic geography theory principles—location, agglomeration, and urbanisation. Secondly, spatial analysis and subsequent clustering integrate companies’ spatial positions, their proximity to other companies, their freight intensity, and the characteristics of the zonal road infrastructure; these clusters are defined as freight traffic zones (FTZs). Thirdly, a functional yet strategic UFI clustering or grouping is proposed to work in an optimised and integrated manner with the FTZs’ opportunities for efficiency and reduced externalities. It is expected that the integrated result of these three components can serve to optimise freight initiatives and road infrastructure from a city governance perspective, reduce freight externalities, and function as a stakeholder cooperation tool from government-led, policy-driven perspectives. This research also identifies and characterises various variables influencing the emergence and existence (planned or organic) of FTZs and shows how these could be incorporated into high-level UPUM processes. Although it is deemed that the principles and methodological approach followed here could be common to urban areas, an example for the Metropolitan Area of the Aburra Valley (MAAV), in Colombia, is presented as an initial case study. Conclusively, this paper introduces a pioneering methodology for integrating UFIs into city or metropolitan governance, offering guidance for policymakers to promote sustainable freight systems.

Despite rail’s growing popularity as a mode of freight transportation due to its role in intermodal transportation and numerous economic and environmental benefits, optimizing all aspects of rail infrastructure use remains a significant challenge. To address this issue, various methods for developing train disruption prediction models have been used. However, these models continue to struggle with accurately predicting short-term arrival delay times, as well as identifying the causes of delays and the expected impact on operations. The lack of information available to operators makes it difficult for them to effectively mitigate the effects of disruptions. The goal of this study is to investigate a set of data-driven models for the short-term prediction of arrival delay time using data from the National Railway Company of Luxembourg of freight rail operations between Bettembourg (Luxembourg) and other nine terminal stations across the EU, and then investigate the effects of the features associated with the arrival delay time. For our dataset, the lightGBM model outperformed other models in predicting the arrival delay time in freight rail operations, with departure delay time, trip distance, and train composition appearing to be the most influential features in predicting the arrival delay time in the short-term. The National Railway Company of Luxembourg can use the short-term prediction model developed in this study as a decision-support system. For example, knowing a train’s arrival delay time allows you to estimate future operational time, providing more support to reduce disruptions and subsequent operational delays via a simple web service.

The importance of analyzing travel mode choice to understand travel behavior in urban areas is crucial in the formulation of mobility policies, especially considering the growing concern to reduce the use of private transport modes in favor of sustainable ones. Although multinomial logit models have been widely used in travel behavior research, machine learning models are becoming an interesting alternative to perform this task, in which tree-based ensemble models, such as random forest and gradient boosting models, have demonstrated superiority in accomplishing this goal, although they both have not been compared using household survey data. This paper compares different logit and machine learning models, with a specific emphasis on gradient boosting, random forest, and multinomial logit models to predict travel mode choice and to identify the determinants in travel behavior in an urban area for three transport modes (public transport, private transport, and walking/bike), using household survey data. Although the methodology is defined following the case and features of the metropolitan area of the Aburrá Valley—Colombia (MAAV), it can be applied to any urban area. The results show that an optimized gradient boosting model is able to predict travel mode choice in an urban area using household survey data, outperforming the other compared models. In addition, travel time, parking type at the destination, the number of motorized vehicles in the household (cars and motorbikes), age, and gender are features that explain the travel mode choice in the MAAV. The optimized gradient boosting model presented in this paper can be employed as a policy tool to study and analyze strategies to promote the reduction of the use of private transport modes in the MAAV and increase the use of more sustainable transport modes.