Sequence classification deals with the task of finding discriminative and concise sequential patterns. To this purpose, many techniques have been proposed, which mainly resort to the use of partial orders to capture the underlying sequences in a database according to the labels. Partial orders, however, pose many limitations, especially on expressiveness, i.e. the aptitude towards capturing certain behavior, and on conciseness, i.e. doing so in a compact and informative way. These limitations can be addressed by using a better representation. In this paper we present the interesting Behavioral Constraint Miner (iBCM), a sequence classification technique that discovers patterns using behavioral constraint templates. The templates comprise a variety of constraints and can express patterns ranging from simple occurrence, to looping and position-based behavior over a sequence. Furthermore, iBCM also captures negative constraints, i.e. absence of particular behavior. The constraints can be discovered by using simple string operations in an efficient way. Finally, deriving the constraints with a window-based approach allows to pinpoint where the constraints hold in a string, and to detect whether patterns are subject to concept drift. Through empirical evaluation, it is shown that iBCM is better capable of classifying sequences more accurately and concisely in a scalable manner.
Due to the unprecedented growth in available data collected by e-learning platforms, including platforms used by MOOC providers, important opportunities arise to structurally use these data for decision making and improvement of the educational offering. Student retention is a strategic task that can be supported by means of automated data-driven dropout prediction. Given the time-based nature of the collected data (user activity), these data can be viewed as sequences, and thus sequence mining presents itself as a fitting set of techniques to automatically extract valuable insights. However, there is a lack of general guidelines for using sequence mining in specific educational settings, as well as little information on how different techniques perform in comparison to each other. We address these limitations with two main contributions. First, we propose a framework for applying sequence classification for dropout prediction in MOOCs. This framework includes two data-driven dropout definitions, the specification of data formatting and preparation tasks, and a blackprint on how to train dropout prediction models at suitable time points in the run of the course. Second, we conduct a benchmarking study of recent and well-performing sequence classification techniques, tested with different parametrizations on 47 real-life datasets from MOOCs, resulting in a comparative assessment of over 18,000 models. Our results provide insight into the performance differences between the techniques and allow us to formulate concrete recommendations towards the choice of suitable hyperparameters that have significant influence on the predictive performance.
Process mining, and in particular process discovery, provides useful tools for extracting process models from event-based data. Nevertheless, certain types of processes are too complex and unstructured to be able to be represented with a start-to-end process model. For such cases, instead of extracting a model from a complete event log, it is interesting to zoom in on some parts of the data and explore behavioral patterns on a local level. Recently, local process model mining has been introduced, which is a technique in-between sequential pattern mining and process discovery. Other process mining methods can also used for mining local patterns, if combined with certain data preprocessing. In this paper, we explore discovery of local patterns in the data representing learning processes. We exploit real-life event logs from JMermaid, a Smart Learning Environment for teaching Information System modeling with built-in feedback functionality. We focus on a specific instance of feedback provided in JMermaid, which is a reminder to simulate the model, and locally explore how students react to this feedback. Additionally, we discuss how to tailor local process model mining to a certain case, in order to avoid the computationally expensive task of discovering all available patterns, by combining it with other techniques for dealing with unstructured data, such as trace clustering and window-based data preprocessing.