Capturing Uncertainty in Unsupervised GPS Trajectory Segmentation Using Bayesian Deep Learning

Over the years, researchers have been studying the effects of weather and context data on transport mode choices. Existing research studies are predominantly designed around travel surveys, but the accuracy of their findings relies on how travelers give accurate and honest answers. The proliferation of smartphones, however, now offers the possibility of utilizing GPS positioning data as an alternative information source, opening the potential to accurately model and better understand factors which influence transport mode choices, compared to travel surveys. The objective of this work is to develop a model to predict the transport mode choices based on GPS trajectories, weather and context data. We use 2671 GPS trajectories from the Geolife GPS trajectories dataset, weather data, such as temperature and air quality, and context data, such as rush hour, day/night time and onetime events, such as the Olympics. In the statistical analysis, we apply both descriptive and statistical models, such as the multinomial logit and probit models. We find that temperature has the most prominent effect among weather conditions. For instance, for temperatures greater than 25 °C, the walking share increases by 27%, and the bike share reduces by 21%, which is line with the results from several survey-based studies. In addition, the evidence of government policy on transport regulation is revealed when the air quality becomes hazardous, as people are encouraged to use environmentally friendly transport mode choices, such as the bike instead of the bus or car, which are known CO2 emitters. Our conclusion is that GPS trajectories can be used as a means to model passenger behavior, e.g. the choice of transport mode, in a quantitative way, which will support transport mode operators and policy makers in their efforts to design and plan the transport mode infrastructure to best suit the passengers’ needs.

In this paper, we propose a Bayesian Deep Learning (BDL) framework to model uncertainty and predict the performance of terahertz (THz) biosensors with a graphene and molybdenum disulfide (MoS2) coating for AML biomarker detection. Although there have been studies on the individual advantage of these 2D materials for biosensing, a comparative analysis taking into account predictive uncertainty is still insufficient. To this end, we have generated a high-fidelity simulation dataset from full-wave EM simulations of DSSRR structures over the 0.1–2.5 THz frequency range. Realistic geometrical and dielectric modifications have been incorporated to mimic bio-sensing conditions. An approach based on a Bayesian Neural Network (BNN) with Monte Carlo dropout was employed for predicting sensitivity, Q-factor, resonance shift, and absorption, along with the estimation of aleatoric, as well as epistemic, uncertainty. Our results demonstrated a trade-off between material types: MoS2 sensors showed higher sensitivity (3548 GHz/RIU) but with a larger prediction uncertainty range of ±118 GHz/RIU; on the other hand, graphene-based sensors exhibited a better spectral resolution (Q = 48.5) and a more reliable QV prediction range of ±42 GHz/RIU. The uncertainty study further revealed that graphene demonstrated a predominance for aleatoric uncertainty (68%), classifying them as predictable physical characteristics, while MoS2 presents a higher epistemic one (55%), indicating sensitivity towards underrepresented design cases. We present a material selection algorithm based on utility that balances sensitivity, resolution, and uncertainty, demonstrating that MoS2 is the best choice for early screening, while graphene is more suitable for high-precision diagnostics. This study offers a scalable and reliable AI framework for quick, uncertainty-aware optimization of THz biosensors, which is directly applicable to clinical diagnostics and 2D-material-based photonic design.

The advancement of spatial information technology (SIT) boosts the location-based services (LBS) and smart urban applications. It brings the public into the era of smart society. Especially, by synthesizing of SIT, smartphones, and urban society, sharing economy mutualizing of goods and services to optimize the resource usage has boomed in recent years. In the transportation sector, transportation network companies (TNC) like Uber and Didi provide flexible and convenient on-demand sharing travel service for the public, thus have significantly changed urban mobility landscape. Taking the TNC vehicle GPS trajectories in Shenzhen as an example, this article presents a data-driven framework to uncover on-demand shared mobility pattern. The travels and stops are extracted from raw GPS trajectories using machine learning. Typical mobility patterns including sharing ridership, travel distance, and spatial communities are revealed. The results suggest that although most sharing travels are concentrated at downtown, a large proportional of LBS-based sharing travels appear at urban sub-centers. The distance of sharing travel follows the exponential distribution with the \(\lambda\) coefficient 0.15. The detected communities indicate the interaction derived from TNC vehicles are constrained by the administrative boundaries. These results suggest the advantages of travel sharing services in places where taxi services are not sufficient. These findings enable us to better understand the online sharing services and provide useful insights on the management of ride-sharing services in the city.

Raw GPS trajectory data are often very large and use up excessive storage space. The efficiency and accuracy of activity patterns analysis or individual–environment interaction modeling using such data may be compromised due to data size and computational needs. Line generalization algorithms may be used to simplify GPS trajectories. However, traditional algorithms focus on geometric characteristics of linear features. Trajectory data may record information beyond location. Examples include time and elevation, and inferred information such as speed, transportation mode, and activities. Effective trajectory simplification should preserve these characteristics in addition to location and orientation of spatial-temporal movement. This paper proposes an Enhanced Douglas–Peucker (EDP) algorithm that implements a set of Enhanced Spatial-Temporal Constraints (ESTC) when simplifying trajectory data. These constraints ensure that the essential properties of a trajectory be preserved through preserving critical points. Further, this study argues that speed profile can uniquely identify a trajectory and thus it can be used to evaluate the effectiveness of a trajectory simplification. The proposed ESTC-EDP simplification method is applied to two examples of GPS trajectory. The results of trajectory simplification are reported and compared with that from traditional DP algorithm. The effectiveness of simplification is evaluated.

Understanding human mobility patterns is of great importance for planning urban and extra-urban spaces and communication infrastructures. The omnipresence of mobile telephony in today's society opens new avenues of discovering the patterns of human mobility by means of analyzing cellular network data. Of particular interest is analyzing passively collected Network Events (NEs) due to their scalability. However, mobility pattern analysis based on network events is challenging because of the coarse granularity of NEs. In this paper, we propose network event-based Bayesian approaches for mobility pattern recognition and reconstruction, mode of transport recognition and modeling the frequent trajectories.

Bayesian deep learning applied to diabetic retinopathy with uncertainty quantification.

In recent years, with the development of science and technology, people have more and more choices for daily travel. However, assisting with various mobile intelligent services by transportation mode detection has become more urgent for the refinement of human activity identification. Although much work has been done on transportation mode detection, accurate and reliable transportation mode detection remains challenging. In this paper, we propose a novel transportation mode detection algorithm, namely T2Trans, based on a temporal convolutional network (i.e., TCN), which employs multiple lightweight sensors integrated into a phone. The feature representation learning of multiple preprocessed sensor data using temporal convolutional networks can improve transportation mode detection accuracy and enhance learning efficiency. Extensive experimental results demonstrated that our algorithm attains a macro F1-score of 86.42% on the real-world SHL dataset and 88.37% on the HTC dataset, with an average accuracy of 86.37% on the SHL dataset and 89.13% on the HTC dataset. Our model can better identify eight transportation modes, including stationary, walking, running, cycling, car, bus, subway, and train, with better transportation mode detection accuracy, and outperform other benchmark algorithms.

Sensitivity analysis of activity scheduling parameters with a parameter optimization framework

Embedded with a wide variety of sensors, such as GPSs, accelerometers, gyroscopes, and microphones, smartphones have become a very useful tool in the context of travel surveys. In this chapter, the authors present an innovative tool to estimate individuals' mobility patterns using an application for smartphones that records GPS and accelerometer data from trips annotated by the user. The authors also present a neural network model for the classification of trips into four transportation modes, based on features extracted from the accelerometer signal. A small sample was collected in Valencia (Spain) to train and evaluate the model. The best classification results were achieved for detecting walking trips (98.2%) and bike rides (99.3%).

Bayesian deep learning is one of the key frameworks employed in handling predictive uncertainty. Variational inference (VI), an extensively used inference method, derives the predictive distributions by Monte Carlo (MC) sampling. The drawback of MC sampling is its extremely high computational cost compared to that of ordinary deep learning. In contrast, the moment propagation (MP)-based approach propagates the output moments of each layer to derive predictive distributions instead of MC sampling. Because of this computational property, it is expected to realize faster inference than MC-based approaches. However, the applicability of the MP-based method in deep models has not been explored sufficiently, even though some studies have demonstrated the effectiveness of MP only in small toy models. One of the reasons is that it is difficult to train deep models by MP because of the large variance in activations. To realize MP in deep models, some normalization layers are required but have not yet been studied. In addition, it is still difficult to design well-calibrated MP-based models, because the effectiveness of MP-based methods under various variational distributions has also not been investigated. In this study, we propose a fast and reliable MP-based Bayesian deep-learning method. First, to train deep-learning models using MP, we introduce a batch normalization layer extended to random variables to prevent increases in the variance of activations. Second, to identify the appropriate variational distribution in MP, we investigate the treatment of moments of several variational distributions and evaluate their uncertainty quality of predictions. Experiments with regression tasks demonstrate that the MP-based method provides qualitatively and quantitatively equivalent predictive performance to MC-based methods regardless of variational distributions. In the classification tasks, we show that we can train MP-based deep models by extended batch normalization. We also show that the MP-based approach realizes 2.0-2.8 times faster inference than the MC-based approach while maintaining the predictive performance. The results of this study can help realize a fast and well-calibrated uncertainty estimation method that can be deployed in a wider range of reliability-aware applications.

The transportation system is the backbone of nation's economy and plays a vital role in the environmental changes in the region. In order to make the transportation system more efficient, the mobility patterns using the data collected from smartphone users can be exploited. For such context-aware systems, transportation mode detection can be of particular interest in different intelligence applications to increase the quality of transportation, traffic safety, and other services. In this paper, various machine learning (ML) techniques are used to classify five different modes of transportation. Specifically, we consider support vector machine (SVM), K-nearest neighbors (KNN), decision tree (DT), bagging, and random forest (RF) algorithms for classification. The performance of all the considered algorithms is compared on the basis of F1 score, accuracy, precision, and recall and tested on freely available TMD dataset.

The data set covers mobility diaries in 6.465 German car users for one week. This data set is based on the work of [1]. The car users noted their mobility behavior in for a whole week. The participants started logging their mobility patterns on different days. In the end, the data was aggregated starting on a Monday. The time of data collection was selected so that there were no holidays in the data. This data for different driver types are found in the CSV-files named DrivingProfiles_*.CSV During the week the participants noted where they went, \eg work, shopping, home, the mode of transportation and the distance traveled. This resulted in mobility diaries in 15 minutes resolution. Each column represents a 15-minute slot. One driver is resembled by to lines. First, her location, second, her distance traveled by car in the 15-minute slot. From this data, we derived parking events with different features found in ParkingEventsData.csv [1] Zumkeller, D., Chlond, B., Ottmann, P., Kagerbauer, M., & Kuhnimhof, T. (2011). Deutsches Mobilitatspanel (MOP)–wissenschaftliche Begleitung und erste Auswertungen. Kurzbericht. Karlsruhe: Institut fur Verkehrswesen, Universitat Karlsruhe. The original data set can be acquired via https://daten.clearingstelle-verkehr.de/192/

This study investigated the influence of accessibility and transportation factors on residential location choices in Kisumu City, Kenya, using a multinomial logistic regression (MLR) framework. Principal component analysis was used to reduce the data into fewer constructs. Three key components, accessibility to relatives and friends, transport services, and social amenities, were operationalized as composite constructs comprising multiple independent variables. The component accessibility to social amenities was comprised of accessibility to shopping centers, sporting facilities, recreational areas, car parks, and modes of private transportation such as motorbikes, tuktuks, and private vehicles. The variable accessibility to transport services was comprised of accessibility to public transport, accessibility to relatives and friends, and the use of motorbikes, tuktuks, and matatus, while the component labeled accessibility to relatives and friends comprised a cluster of independent variables, including accessibility to relatives and friends, walking as a mode of transport, accessibility to bicycles, matatus (public minibuses), and private vehicles. The high-income residential category served as the reference group for comparative analysis. Results from likelihood ratio tests confirmed that all three components significantly affect residential decisions (p = .000). The 'accessibility to relatives and friends' component was strongly associated with preferences for low- and middle-income neighborhoods, with odds ratios of 8.429 and 15.357, respectively. Similarly, the 'accessibility to transport services' component increased the likelihood of selecting low- and middle-income areas by 386% and 500%, respectively, relative to high-income neighborhoods. In contrast, the 'accessibility to social amenities' component showed a negative association with low- and middle-income residential choices, with odds ratios of 0.138 and 0.120, indicating that concern for social amenities significantly favored high-income neighborhood selection. These findings suggest that while social and transport accessibility is critical for lower-income households, access to amenities is a stronger determinant for high-income residents. The results reflect broader socio-economic dynamics, where affordability constraints and limited infrastructure in low-income areas lead residents to prioritize social proximity and walkability over formal amenities. The study contributes to urban planning discourse by emphasizing the need for inclusive housing policies that account for both social networks and mobility patterns in rapidly urbanizing cities.

IoT data with timestamps are often found with outliers, such as GPS trajectories or sensor readings. While existing systems mostly focus on detecting temporal outliers without explanations and repairs, a decision maker may be more interested in the cause of the outlier appearance such that subsequent actions would be taken, e.g., cleaning unreliable readings or repairing broken devices or adopting a strategy for data repairs. Such outlier detection, explanation, and repairs are expected to be performed in either offline (batch) or online modes (over streaming IoT data with timestamps). In this work, we present TsClean, a new prototype system for detecting and repairing outliers with explanations over IoT data. The framework defines uniform profiles to explain the outliers detected by various algorithms, including the outliers with variant time intervals, and take approaches to repair outliers. Both batch and streaming processing are supported in a uniform framework. In particular, by varying the block size, it provides a tradeoff between computing the accurate results and approximating with efficient incremental computation. In this article, we present several case studies of applying TsClean in industry, e.g., how this framework works in detecting and repairing outliers over excavator water temperature data, and how to get reasonable explanations and repairs for the detected outliers in tracking excavators.

The results of a wide survey with 10 000 participants in 20 European cities suggest that mobility patterns in 2021 were still affected by the Covid-19 pandemic. The survey explores the factors that influence the respondents’ choices as regards trip generation, distribution and modal choice. The behavioural changes resulting from mobility restrictions in 2020 and the adoption of technological alternatives such as teleworking, video-conferencing, or online shopping limited the number of trips, but the gradual return to normality is also leading to a rebound in transport activity. The results reveal an increased use of cars for daily urban mobility, which may be sustained in the future. The main reason seems to be an increased level of users’ risk aversion and their avoidance of public transport or other shared modes. Such a trend can be also alarming as regards the future prospects of emerging technologies and business models, e.g. micro-mobility or Mobility-as-a-Service applications.