<![CDATA[The rapid growth of mobility data from GPS trajectories offers unprecedented opportunities to gain deep insights into human mobility behavior, with significant implications for urban planning, traffic management, public transportation optimization, emergency response, and smart city development. However, a key challenge lies in transforming raw GPS trajectory data, consisting of sequences of coordinates and timestamps, into meaningful, context-rich information that can support analysis and decision making. This study proposes a semi-supervised framework to enhance the contextual and semantic understanding of journeys, using Grab Jakarta GPS trajectory data as a case study. The framework involves extracting origin-destination pairs, augmenting the data with temporal (day, time) and spatial (postal code, land use) contexts through public datasets, assigning cluster labels to characterize groups of journeys, analyzing mobility patterns, and ultimately predicting trip destinations. Origin-destination clustering, performed using the DBSCAN algorithm, identified five meaningful clusters, achieving the highest silhouette score of 0.56 with epsilon = 7.0 and min_samples = 5. Subsequently, a regression-based prediction model was developed, employing nine algorithms, including three deep learning approaches. The LSTM model demonstrated the best performance, yielding a mean squared error of 0.0053 and a coefficient of determination (R²) of 86.20% in predicting trip destinations. These findings highlight the potential of integrating spatial-temporal enrichment and machine learning to derive actionable insights from GPS trajectory data.]]>
Copyrights © 2025
