Modeling and predicting data for business intelligence
MetadataShow full item record
Business intelligence is an area where data and actionable information can be analyzed and provided to make more informed business actions. In general, any technique that contributes to better business decisions can be categorized in business intelligence techniques. Particularly, business process management is a subarea that focuses on improving business performance by managing and optimizing business processes; management data mining is a subarea that applies data mining techniques to gain better business performances. In business process management, a business process is a collection of relevant, structured activities or tasks that produce specific services or products for certain business goals. Business process modeling refers to the activities of representing intra or inter-organization processes, such that the current processes can be analyzed or improved. While abundant business process modeling techniques and their associated analyses have been proposed to capture different aspects of business processes, modern business processes can be very complicated such that many properties, such as performance optimization and evaluation, still cannot be accurately described and understood. Management data mining refers to applying data mining techniques in multiple domains of business managements, e.g., supply chain management, marketing analysis. Typical research topics include building models to provide feedbacks for skewing supply chain policies or marketing strategies. Traditional research tend to build generic models given specific scenarios, that are argued to easily cause inaccuracies with more granular disturbances. My thesis focuses on approaches handling the challenges in business process optimization and evaluation and its associated data analysis. Specifically, I propose a data-centric technique for modeling composite business activities by including components of data, human actors, and atomic activities. Furthermore, I explore this technique in two major dimensions. First, by applying this technique in work ow-based business processes, I explore the possibility of reconstructing these processes by modifying the execution order of business activities, and develop efficient algorithms to approach optimal temporal performance for data-centric business processes. Second, I build a symbolic process generator to stochastically generate symbolic data-centric business processes that can be used to analyze their properties and evaluate optimization approaches according to end-users' specification. Moreover, I zoom in a granular data type of inventory management process and build data mining models to forecast it. The major contributions of my thesis include: 1) proposing a data-centric business process modeling technique that emphasizes business artifacts compared with traditional workflow-based modeling techniques; 2) developing approaches to optimize the temporal performance of the data-centric business processes; 3) applying my symbolic process generator so that data-centric business processes can be simulated and provided with quick and inexpensive analyses. 4) building data mining models for forecasting inventory shipments based on real-world data sets.