The new study, published in the International Journal of Data Science and Analytics by Akilu Rilwan Muhammad, Ana Aguiar, and João Mendes-Moreira, presents a class-aware classification framework called Class-subspace. Its central innovation lies in borrowing an idea from cooperative game theory: the Shapley value. Originally developed to fairly divide a payout among players in a coalition, Shapley values have become one of the most trusted tools for explaining machine learning predictions through the SHAP method, short for SHapley Additive exPlanations. Rather than using SHAP merely to explain a model after the fact, the Porto team flips the workflow, using SHAP attributions to actively construct the model itself.

The reasoning is deceptively simple. Conventional transportation mode detection pipelines apply a single, globally selected feature set across every travel mode, optimising one subset of variables for the entire multi-class problem. But the researchers found that the features most relevant for distinguishing a bus journey were often sub-optimal for identifying a car trip. A single global feature set, they argue, necessarily compromises performance for certain classes. Class-subspace instead trains an initial classifier, computes mean absolute SHAP values for each feature and each class, ranks the features by their contribution to each mode’s prediction, and retains the minimal set of features that explains a cumulative threshold of 60 percent of each class’s relevance. Each transportation mode thus acquires its own compact feature subspace, and a specialised base learner is trained for every mode before their votes are aggregated into a final decision.

The technical pipeline behind the framework is substantial. The team worked with two real-world GPS trajectory datasets: SenseMyFEUP, collected from 227 participants in Porto, Portugal, using an Android application that sampled location roughly once per second, and the well-known GeoLife dataset from Microsoft Research Asia, gathered from 182 participants in Beijing over nearly five years. Rigorous preprocessing was essential. The researchers chained GPS points into trips using a 30-minute gap threshold, filtered out anomalous sessions that continued recording without movement, applied a spatial filter to keep only city-scale journeys, and capped implausible walking trips at eight kilometres. Segments shorter than 50 metres were discarded, removing nearly 39 percent of segments in the Porto data and about 14 percent in GeoLife.

Feature engineering drew on both the time and frequency domains. For each 100-step instance, the team computed statistics of speed, acceleration, jerk, and bearing rate, capturing the kinematic character of motion. They then applied the Fast Fourier Transform to convert these signals into the frequency domain, hypothesising that cyclic human movements such as walking or pedalling would produce distinctive high-frequency components, while smooth mechanical travel by car or bus would concentrate energy at low frequencies. In total, 80 features per instance were generated, with frequency-domain information encoded through the indices of the top ten spectral components rather than their magnitudes, capturing which frequency bands are active for each mode rather than how much energy they carry.

Evaluation was deliberately conservative. Instead of the random data splits common in the field, which can leak future patterns into training and inflate accuracy, the team used temporal splits, training on earlier data and testing on later periods. Against baselines including Sequential Forward Floating Selection, Mutual Information filtering, and the Boruta algorithm, Class-subspace achieved weighted ROC-AUC scores of up to 78.8 percent on the Porto dataset and 89.5 percent on GeoLife using the XGBoost classifier, performing comparably to established global feature selection methods. A formal statistical comparison using the Friedman test and Bonferroni-Dunn post hoc analysis confirmed that no baseline differed significantly from the proposed method.

What the framework sacrifices in raw accuracy gains, it returns in interpretability. An analysis of feature overlap across class-specific subspaces, measured with the Jaccard similarity index, revealed moderate overlap and a small universal core of speed-based statistics, including maximum, mean, standard deviation, and 85th-percentile speed, that appeared in every configuration. Beyond that shared core, each mode carried its own signature. Walking was captured almost entirely by speed statistics, consistent with its narrow and stable range. Cycling added bearing rate features reflecting frequent directional changes. Bus detection drew on jerk and bearing rate, echoing stop-and-go dynamics. Metro uniquely selected bearing rate distributions and acceleration frequency components, matching the confined, scheduled vibration patterns of rail transit. On the Porto data with random forest, only 7 of 32 distinct features were common to all classes, while 9 served exclusively as metro signatures.

The authors are candid about limitations. Bus and metro classes remained difficult on the Porto dataset, a challenge they attribute to the joint effect of class imbalance and class overlap rather than to feature selection itself, pointing to their earlier work characterising these issues. The computational cost of Shapley values is a known constraint, and the sensitivity of performance to the contribution threshold varied between datasets, with GeoLife remarkably stable across thresholds while the Porto data peaked sharply at 0.60. Future work, they suggest, could explore adaptive threshold selection through cross-validated grid search, Bayesian optimisation, or information-theoretic criteria, and extend the class-subspace concept to other multi-class domains.

The broader implications reach well beyond transportation research. As cities worldwide lean on crowdsensed mobility data for planning, emissions modelling, and infrastructure investment, the trustworthiness of the underlying classification models matters enormously. Class-subspace demonstrates that per-class interpretability need not come at the price of predictive power: a model can match the performance of globally optimised baselines while explicitly revealing which evidence supports each decision. In a field where reported accuracies above 90 percent sometimes rest on leaky evaluation protocols, the Porto team’s insistence on temporal splits and statistical rigor sets a standard. Their message to the machine learning community is clear: when classes are fundamentally different, stop forcing them to share the same lens, and let game theory hand each one its own.

Subject of Research: SHAP-driven class-aware machine learning for transportation mode detection from GPS trajectory data

Article Title: Class-subspace learning: a SHAP-driven framework for class-aware transportation mode detection

Article References: Muhammad, A. R., Aguiar, A., & Mendes-Moreira, J. (2026). Class-subspace learning: a SHAP-driven framework for class-aware transportation mode detection. International Journal of Data Science and Analytics, 22(1), Article 302. https://doi.org/10.1007/s41060-026-01263-x

Image Credits: AI Generated

DOI: 10.1007/s41060-026-01263-x

Keywords: transportation mode detection, SHAP values, machine learning, GPS trajectories, feature selection, Shapley values, urban mobility, smart mobility, interpretability, XGBoost, Fast Fourier Transform, class-aware classification

The research, led by Milad Rajaei with Usman Safder, Sarah Cotterill, and Recep Kaan Dereli, presents a city-scale framework that simulates water demand in near real time by explicitly tracking how people move through a city. The work builds on SIMDEUM, a well-established stochastic end-use model originally developed in the Netherlands, which represents water consumption as a stream of random pulses—each toilet flush, shower, tap use, or dishwasher cycle—with timing, duration, and flow rate drawn from probability distributions derived from empirical observations of occupant behaviour. SIMDEUM has proven remarkably capable of reproducing realistic household demand patterns at high temporal resolution, and it has been extended to offices, hotels, nursing homes, and other non-residential buildings by dividing each building into functional rooms with their own appliances and users. But the model has always carried a fundamental weakness: it assumes occupancy is either static or averaged, which in dynamic urban environments can become the dominant source of error.

The importance of occupancy is not in doubt. Sensitivity analyses of stochastic residential demand models have found Spearman’s rank correlation coefficients between occupancy and both peak and average demand ranging from 0.99 to 0.995—an almost perfect relationship. Field experiments reinforce the point: researchers who installed flush counters on 119 toilets across seven university campus buildings demonstrated a strong, direct link between toilet water use and the number of people present. The COVID-19 pandemic made the consequences of ignoring this relationship vividly clear, as commuting collapsed, workplaces emptied, and hygiene practices intensified, producing higher residential demand alongside sharply reduced commercial consumption. Models grounded in static occupancy assumptions simply could not see these shifts coming.

The researchers’ insight was to recognise that the data needed to model dynamic occupancy already exists in aggregate form. Google Popular Times indicators describe how busy non-residential locations are at any given moment, derived from aggregated and anonymised smartphone location data, expressed on a relative scale from 0 to 100 compared with a location’s typical peak activity. The framework begins by collecting these signals at five-minute intervals through automated web scraping for every non-residential building in a study area. Where live data are available and pass quality checks, they are used directly; where they are not, the model descends through a careful hierarchy of fallbacks—historical average patterns for the same day of the week, then representative occupancy profiles derived from K-means clustering of tens of thousands of profiles collected nationwide, then literature-based profiles from U.S. Department of Energy reference buildings, adjusted with local correction factors for seasonal effects such as school terms and hotel occupancy statistics.

Converting relative busyness into absolute occupant numbers requires a further step: each Popular Times value is multiplied by the estimated capacity of the building, calculated by dividing floor area by occupancy load factors taken from building design and fire safety guidelines, and scaled to reflect normal operation rather than maximum permitted crowding. Data quality proved to be a genuine challenge. Across a one-month collection period in September 2024, 58 percent of live samples were classified as invalid under the study’s quality-control rules, which flagged suspicious sudden drops in occupancy that persisted briefly before abruptly returning to normal—patterns unlikely to represent real activity. Days with insufficient valid data were replaced wholesale with historical averages, while shorter gaps were filled by linear interpolation. The prevalence of anomalies, particularly at low-traffic locations, underscores that crowdsourced occupancy data is useful but demands rigorous preprocessing.

The most conceptually ambitious element of the framework is its treatment of residential occupancy, for which no direct crowdsourced signal exists. Rather than relying on census averages, the model infers where people are at home by tracking population movements. Drawing on two classic theories of human mobility—Zipf’s gravity model, which holds that movement likelihood rises with population and falls with distance, and Stouffer’s intervening opportunities model, which assumes people choose the nearest destination that satisfies their needs—the framework constructs a trip probability matrix at each time step. When non-residential occupancy rises, the corresponding number of people is drawn probabilistically from residential areas weighted by their populations and the distribution of nearby opportunities; when occupancy falls, people return to their original home areas. A tourist population, calibrated from national accommodation and tourism statistics, handles movements associated with hotels and nightlife, while a separate commuter population accounts for people travelling into the study area from outside.

These time-varying occupancy estimates then feed directly into a modified SIMDEUM model running at one-minute resolution in MATLAB. At each time step, the probability of a water-use event for each end-use is calculated from the occupancy, the per-person frequency of use, and a diurnal timing factor reflecting behavioural routines. A random draw is compared with this probability to decide whether an event occurs, and if so, its flow rate and duration determine the volume consumed, with each end-use temporarily locked during an event to prevent overlap. For occupancy-dependent end-uses such as toilet flushing and hand washing, the occupancy term drives the calculation; for scheduled activities such as office cleaning, occupancy is effectively set aside so that only frequency and timing matter.

Applied to Sligo, a coastal town of roughly 20,000 people in northwest Ireland, the framework simulated an entire month of city-scale demand. Residential consumption came out at approximately 129 litres per person per day—closely matching the metered benchmark of about 312 litres per household per day reported for Sligo—and the simulated end-use breakdown, with toilets accounting for 28 percent of consumption, showers 24 percent, and kitchen taps 21 percent, differed by no more than two percentage points from published values for Irish households. The temporal patterns behaved as one would expect: a pronounced morning peak between 6:00 and 9:00 a.m. on weekdays driven by showering and breakfast routines, a delayed peak on weekends, a midday dip as residents left for work or school, and an evening recovery as people returned home. Non-residential demand told equally coherent stories—restaurants showed sharp peaks aligned with mealtimes, food retail displayed the steadier profile of continuous cleaning and toilet use, and office buildings peaked at the start of the working day before collapsing after closure.

The study is candid about its limitations. The default SIMDEUM parameters derive from Dutch household data and may not transfer cleanly to Irish conditions—simulated restaurant water use of 6.2 litres per square metre per day diverged from the 2.48 litres reported by Irish Water for comparable commercial premises, a discrepancy the authors attribute primarily to uncalibrated appliance frequencies, durations, and flow rates. The conversion of relative busyness into absolute occupant counts also requires independent validation against footfall sensors or building occupancy systems. Nevertheless, the results demonstrate something genuinely significant: a scalable, transferable route to high-resolution water demand modelling that requires no dense smart metering infrastructure whatsoever. By resolving demand at the level of individual end-uses across entire cities, the framework opens the door to demand-responsive pumping schedules, energy-aware operation of distribution networks, and scenario testing for planners—capabilities that could meaningfully reduce the energy intensity and emissions of the urban water cycle, one flush at a time.

Subject of Research: A real-time, city-scale water demand modelling framework that integrates urban mobility data from Google Popular Times with a stochastic end-use water demand model.

Article Title: A framework for real-time water demand modelling at city scale based on urban mobility

Article References: Rajaei, M., Safder, U., Cotterill, S., & Dereli, R. K. (2026). A framework for real-time water demand modelling at city scale based on urban mobility. Energy Reports, 16, Article 109700. https://doi.org/10.1016/j.egyr.2026.109700

Image Credits: AI Generated

DOI: 10.1016/j.egyr.2026.109700

Keywords: water demand modelling, urban mobility, Google Popular Times, SIMDEUM, smart meters, occupancy, water-energy nexus, end-use model, stochastic simulation, urban water systems, Sligo Ireland, energy efficiency

The motivation stems from a persistent and costly bottleneck. Cross-domain research and development in urban mobility—where batteries, artificial intelligence, vehicle-to-everything communication, and logistics systems must converge—has long been stalled by disciplinary silos. Traditional foresight methods such as expert panels, technology roadmaps, and bibliometric analyses can identify promising trends, but they rarely translate those trends into actionable, structured collaboration plans that pair the right industrial innovators with the right academic partners. Lee’s framework is designed to close that gap by converting patent-based convergence signals directly into industry–academia R&D proposals.

Technically, the system operates in several tightly coupled stages. First, a Bayesian long short-term memory (LSTM) network performs walk-forward signal detection on patent classification data, identifying high-potential technology pairs whose convergence trajectories suggest fertile ground for joint research. The walk-forward design is deliberately conservative: the model is trained only on earlier periods and tested on immediately following held-out periods, with no access to target data during training. In the short-window setting—training on 2000–2014 and predicting 2015–2019—the Bayesian LSTM achieved perfect directional accuracy on six validated technology pairs, outperforming four transparent baseline models under identical conditions.

Once promising technology pairs are identified, the framework builds expert personas directly from inventor records in the patent corpus. Rather than relying on simple patent counts, Lee employs a Linear Weighted Score that recency-weights each inventor’s contributions: a patent filed in 2024 receives roughly 25 times the weight of one filed in 2000. This ensures that simulated experts reflect the current innovation frontier rather than historical output. Sensitivity checks confirmed the robustness of this approach—replacing the weighted score with raw counts or exponential weighting shifted top-10 expert rankings by at most two positions in most cases. The distribution analysis also revealed a striking concentration of expertise: both productivity and influence are heavily right-skewed, with high-impact innovators being exceptionally rare.

The heart of the system is a three-phase dialogue orchestrated through LangGraph, a framework for coordinating multi-agent workflows. In Phase 1, two industry expert agents—constructed from real inventor data, such as specialists in AI-driven smart parking systems and notification control technologies—hold a virtual convergence meeting under one of three facilitation strategies: Consensus-Driven, Greedy-Exploitation, or Exploratory-Brainstorming. The goal is to identify the single most critical academic research field needed to enable a proposed technology fusion. In Phase 2, the system queries ArXiv to discover a suitable academic collaborator, synthesizing the literature into a detailed persona. In Phase 3, the industry and academic agents convene to produce a structured joint R&D plan with quarterly milestones, role divisions, and expected outcomes, output as a parsable JSON meeting log.

The results from 30 capability-demonstration runs were striking. ArXiv retrieval succeeded on the first or second query in all 30 runs, with independent raters scoring persona-to-knowledge-gap alignment at a mean of 4.1 out of 5.0. Every refined proposal incorporated at least one frontier AI paradigm—foundation models, federated learning, quantum machine learning, or agentic intelligence—compared with none at the initial convergence stage, illustrating what Lee terms the candidate synthesis effect. Only one run fell below the quality threshold, and a structural analysis traced the failure to a mismatch between the Greedy-Exploitation strategy and the specific technology pair, producing a proposal with too few milestones to be actionable.

What distinguishes the study methodologically is its Two-Track evaluation architecture, designed to confront one of the most serious criticisms of LLM-based research: the ‘LLM-as-judge’ circularity problem, in which the same family of generative models both produces and evaluates the output. Track A demonstrates the framework’s generative ceiling using GPT-4-class models. Track B, the primary validity evidence, employs six non-generative discriminative scorers—none sharing computational lineage with the generative models—to produce a Collaborative Quality Score. This design substantially reduces evaluator–generator circularity, yielding a pipeline advantage of Δ = +0.275 with a large effect size (d = 2.227) over strong baselines, including a cross-family comparison against Alibaba’s Qwen2.5-3B under both chain-of-thought and direct prompting conditions.

Human validation reinforced the automated findings. Three independent domain experts—an AI/ML professor, a senior ITS/V2X research engineer, and an innovation policy specialist—blind-rated 24 proposals using a 13-item rubric, achieving an intraclass correlation of 0.918, indicating strong inter-rater reliability. An FDR-corrected ablation analysis at 35 runs per condition showed that removing academic integration or LSTM forecasting produced large-effect degradations in proposal quality, while removing personas caused medium-effect degradation—evidence that the pipeline’s modules work synergistically rather than redundantly. Notably, a calibration gap emerged: automated scores were systematically higher than human ratings, and sub-dimension correlations between proxy scorers and holistic human judgments were low, an honest limitation the study documents in detail.

Lee is candid about the framework’s boundaries. Signal detection operates over relatively short windows, automated scoring remains proxy-level, and real-world feasibility of the generated proposals has not yet been confirmed—no simulated collaboration has been executed by actual laboratories or companies. The generalization check comparing centrality-selected versus randomly sampled technology pairs is explicitly flagged as confounded and non-confirmatory, with the author calling for replication with at least 20 pairs per group under a common evaluation model. These caveats, far from undermining the work, reflect a design science ethos in which transparency about limitations accompanies every claim.

The implications nonetheless extend well beyond urban mobility. If patent data can seed credible expert personas, if retrieval systems can locate genuine academic counterparts, and if orchestrated multi-agent dialogue can produce structured, evaluable R&D roadmaps, then the same architecture could be applied to energy transition, biotechnology, semiconductor design, or any domain where convergence across disciplinary boundaries determines the pace of innovation. All datasets and simulation outputs have been released in a public repository, inviting replication and extension. As generative AI matures from a tool for drafting text into infrastructure for planning discovery itself, this study offers a rigorous, carefully validated template for what simulated innovation ecosystems might look like—and a sober reminder that trustworthy evaluation, not generation alone, is what will make them useful.

Subject of Research: A data-grounded multi-agent large language model framework for planning industry–academia R&D collaboration in urban mobility using patent-based technology convergence signals.

Article Title: Simulating the Future of Innovation: A Data-Grounded, Multi-Agent LLM Framework for R&D Collaboration Planning in Urban Mobility

Article References: Lee, Y.-J. (2026). Simulating the Future of Innovation: A Data-Grounded, Multi-Agent LLM Framework for R&D Collaboration Planning in Urban Mobility. Information Systems Frontiers. https://doi.org/10.1007/s10796-026-10797-1

Image Credits: AI Generated

DOI: 10.1007/s10796-026-10797-1

Keywords: urban mobility, multi-agent LLM, R&D collaboration, patent analysis, technology convergence, Bayesian LSTM, innovation simulation, LangGraph, ArXiv retrieval, expert personas, large language models, technology forecasting

The research, led by M Sana Ullah Khan and Fumiko Ito of Tokyo Metropolitan University, tackles a persistent blind spot in how scientists and operators understand bike-share networks. Most previous studies have relied on trip counts and origin-destination flows to describe demand, which are useful for measuring overall ridership but say little about the short-term operational stress that individual stations experience. What matters for service reliability, the authors argue, is whether bicycles are accumulating or depleting at a station over short intervals, creating empty docks or full docks that force users to search elsewhere.

To capture this, the team used a metric called net bike change: simply, the difference in available bicycles at a station between two consecutive fifteen-minute snapshots. A positive value means more bikes were returned than taken, marking the station as a net destination; a negative value means departures dominated, marking it a net origin. The researchers selected the fifteen-minute resolution after pilot testing showed that finer intervals were dominated by noise from single-bike movements while thirty-minute intervals smoothed away the crucial timing of morning and evening transitions. Crucially, the measure comes from publicly available station-availability feeds, requiring no proprietary trip records, making the approach cheap, reproducible, and transferable to cities where origin-destination data simply do not exist.

The dataset covered 3,207 stations from Tokyo’s two dominant dock-based systems, Hello Cycling and Docomo Bike Share, observed across three seasonal windows: October 2023, January 2024, and May 2024. Each window spanned seven consecutive days from 07:00 to 23:00 at fifteen-minute intervals, and only stations present in all three seasons were retained to ensure comparability. From these records the researchers built temporal imbalance profiles for every station, separate profiles for weekdays and weekends, each containing sixty-four values representing the day between 07:15 and 23:00.

Because such profiles are high-dimensional and highly correlated, the team first applied Principal Component Analysis to compress each station’s behaviour into a small set of interpretable temporal contrasts. A scree plot pointed to four principal components, which together explained about nineteen percent of total variance, a modest share that reflects the noise inherent in station-level dynamics but sufficient to preserve the dominant morning-versus-late-day and weekday-weekend patterns. Hierarchical clustering using Ward’s method, which merges stations while minimising within-group variance, then produced a strikingly clean three-cluster solution, validated by a silhouette score of 0.767 and confirmed as robust across alternative specifications.

The three clusters turned out to have vivid and intuitive meanings. The first group, commuter-destination stations, shows a sharp weekday morning inflow between roughly 07:15 and 09:00 as bikes flood into commercial and employment areas, followed by a pronounced evening outflow around 19:00 as workers ride away. The second group, commuter-origin stations, displays the mirror image: strong morning outflow from residential and mixed inner-city neighbourhoods, then steady evening inflow as bikes return home. The third group comprises balanced or low-signature stations with nearly flat profiles around zero across both weekdays and weekends, indicating either evenly matched arrivals and departures or generally low activity.

Perhaps the most consequential finding is how stable these roles proved across seasons. Tracking cluster membership for each individual station across autumn, winter, and spring, the researchers found that 86.4 percent of stations stayed in the same role in all three seasons, and pairwise agreement between seasons ranged from 89.3 to 93.1 percent. Persistence was strongest for the balanced role, which retained between 96.8 and 99.2 percent of its stations across any seasonal pair. The commuter roles were somewhat more season-sensitive, with the commuter-destination cluster shrinking from 209 stations in autumn to just 15 in spring as some stations drifted into the balanced category, but the core weekday structure held firm. Weekend profiles, by contrast, were consistently flatter and more dispersed, confirming that discretionary leisure use forms a secondary, less predictable layer atop the rigid weekday skeleton.

The spatial mapping of these roles reveals Tokyo’s urban anatomy with unusual clarity. Commuter-destination stations concentrate in the commercial core and bay-side employment districts served by major rail terminals, with 72.7 percent located in commercial land-use areas and, remarkably, 100 percent lying within 800 metres of a train station, the distance of roughly a ten-minute walk. Commuter-origin stations cluster in inner-city sub-centres and mixed-use zones, showing the most heterogeneous land-use profile with the highest industrial share. Balanced stations spread across residential outer wards and the western Tama municipalities, where rail stations are more widely spaced. Statistical tests confirmed that cluster membership was significantly associated with land-use type and train-station proximity, but not with bus-stop proximity, suggesting that rail accessibility, not bus coverage, is what structurally shapes the bike-share network.

The practical implications are direct. Rebalancing operations, the costly business of trucking bicycles from full stations to empty ones, can be organised according to station role and time of day rather than applied uniformly. Stations with morning inflow need empty dock capacity cleared in advance of the peak; stations with morning outflow need bikes pre-positioned before commuters arrive. Balanced stations may require far less frequent intervention, freeing resources for the hotspots where recurrent directional pressure is strongest. More broadly, because the method depends only on publicly available availability data, operators in cities worldwide could replicate the classification without access to proprietary trip logs, giving transit planners a role-based tool for integrating shared bicycles into last-mile networks.

The study also extends previous station-typology research by demonstrating, in a dense rail-oriented metropolis, that the functional meaning of station groups persists across multiple seasons rather than being re-formed each time. The authors caution that net bike change is an indirect proxy for demand, partly shaped by operator rebalancing and dock-capacity constraints, and that their analysis rests on seasonal snapshots rather than a full-year panel. Future work, they suggest, should incorporate weather and event effects, cycling infrastructure, capacity-normalised measures, and comparisons with cities of different urban forms. But the central message stands: in Tokyo, the shared bicycle is not an independent mode but a finely tuned appendage of the railway system, and its daily ebb and flow writes the commuting geography of the city in fifteen-minute installments.

Subject of Research: Classification of bicycle sharing station functional roles in Tokyo using temporal imbalance profiles derived from high-frequency station-availability data.

Article Title: Classifying bicycle sharing station roles using temporal imbalance profiles in Tokyo

Article References: Classifying bicycle sharing station roles using temporal imbalance profiles in Tokyo. (n.d.). https://doi.org/10.1007/s44327-026-00353-6

Image Credits: AI Generated

DOI: 10.1007/s44327-026-00353-6

Keywords: bike sharing, Tokyo, temporal imbalance, net bike change, principal component analysis, hierarchical clustering, commuter stations, last-mile connectivity, transit-oriented development, urban mobility, rail integration, rebalancing