Intelligent transportation systems (ITS) have garnered considerable attention for providing efficient traffic management solutions. Traffic flow forecasting is a crucial component of it which serves as the foundation for various state-of-the-art deep learning approaches. Initially, researchers recognized that significant temporal changes from traffic flow data for modelling. However, as researchers delved deeper into the underlying correlations within traffic flow data, they discovered that spatial information from the road network also plays a crucial role in accurate forecasting. Consequently, deep learning methods that incorporate Spatio-temporal representation have been employed to address traffic flow forecasting. Although recent solutions to this problem are impressive, it is essential to discuss the reasoning behind the architecture of the model. The expression of each feature relies on selecting appropriate models for feature extraction and designing architectures that minimize information loss during modeling. In this thesis, the work focuses on graph-based Spatio-temporal feature engineering. The experiments are divided into two parts: 1). explores the efficient architecture for expressing spatial-temporal information by considering both different sequential modelling approaches. 2). Based on the result obtained, the second experiment focuses on multi- scale modelling to capture informative Spatio-temporal feature. We propose a model that incorporates sequential modeling and captures multi-scale Spatiotemporal semantics by employing residual connections in different hierarchy. We validate our model using three datasets, each containing varying information for extraction. Taking into account the dataset characteristics and the model structure, our model outperforms the baselines and state-of-the-art models. The experimental results indicate that the performance of sequential modeling and multi-scale semantics, combined with thoughtful model design, significantly contribute to the overall forecasting performance. Furthermore, our work serves as inspiration for expressive data mining methods that rely on appropriate feature extraction models and architecture design, taking into consideration the information content within the dataset.

Safe driving is a topic of multiple factors where the road surface condition is one. Knowledge about the road status can for instance indicate whether it is risk for low friction and thereby help increase the safety in traffic. The ambient temperature is an important factor when determining the road surface condition and is therefore in focus. This work evaluates different methods of data fusion to estimate the ambient temperature at road segments. Data from vehicles are used during the temperature estimation process while measurements from weather stations are used for evaluation. Both temporal and spatial dependencies are examined through different models to predict how the temperature will evolve over time. The proposed Kalman filters are able to both interpolate in road segments where many observations are available and to extrapolate to road segments with no or only a few observations. The results show that interpolation leads to an average error of 0.5 degrees during winter when the temperature varies around five degrees day to night. Furthermore, the average error increases to two degrees during springtime when the temperature instead varies about fifteen degrees per day. It is shown that the risk of large estimation error is high when there are no observations from vehicles. As a separate result, it has been noted that the weather stations have a bias compared to the measurements from the cars.

More than ever, businesses heavily rely on IT service delivery to meet their current and frequently changing business requirements. Optimizing the quality of service delivery improves customer satisfaction and continues to be a critical driver for business growth. The routine maintenance procedure plays a key function in IT service management, which typically involves problem detection, determination and resolution for the service infrastructure. Many IT Service Providers adopt partial automation for incident diagnosis and resolution where the operation of the system administrators and automation operation are intertwined. Often the system administrators' roles are limited to helping triage tickets to the processing teams for problem resolving. The processing teams are responsible to perform a complex root cause analysis, providing the system statistics, event and ticket data. A large scale of system statistics, event and ticket data aggravate the burden of problem diagnosis on both the system administrators and the processing teams during routine maintenance procedures. Alleviating human efforts involved in IT service management dictates intelligent and efficient solutions to maximize the automation of routine maintenance procedures. Three research directions are identified and considered to be helpful for IT service management optimization: (1) Automatically determine problem categories according to the symptom description in a ticket; (2) Intelligently discover interesting temporal patterns from system events; (3) Instantly identify temporal dependencies among system performance statistics data. Provided with ticket, event, and system performance statistics data, the three directions can be effectively addressed with a data-driven solution. The quality of IT service delivery can be improved in an efficient and effective way. The dissertation addresses the research topics outlined above. Concretely, we design and develop data-driven solutions to help system administrators better manage the system and alleviate the human efforts involved in IT Service management, including (1) a knowledge guided hierarchical multi-label classification method for IT problem category determination based on both the symptom description in a ticket and the domain knowledge from the system administrators; (2) an efficient expectation maximization approach for temporal event pattern discovery based on a parametric model; (3) an online inference on time-varying temporal dependency discovery from large-scale time series data.