Search results

Process mining helps organizations improve their business processes in today’s data-rich environment. However, these processes can change over time due to factors like policy shifts or process trends, impacting model performance. This study examines process behavior in event logs and uses machine learning to detect concept drift.

The trace clustering and change mining techniques have been implemented on two processes, namely loan payment and temporary identity creation, to detect drifts. We use the bag-of-activities and edit distance methods, along with K-Mode and agglomerative hierarchical clustering techniques.

This study makes two important findings: trace clustering is a popular choice for detecting drifts, and the bag-of-activities method using K-Mode clustering and hamming distance proved highly effective at spotting drifts in various event logs. It also identifies different types of drifts occurring simultaneously in the processes.

The drifts discovered in different processes provide a real-world example of concept drift in the domains of loans and university administrations. This contributes to improving operational efficiency and overall organizational performance based on these detected drifts and assists in enhancing the process design.

This study is the first to employ a hybrid method of trace clustering and change mining to detect process changes. It is also the first to simultaneously detect sudden and recurring drift in the field of trace clustering in process mining. Furthermore, it stands out for investigating and comparing the performance of multiple clustering methods, in contrast to prior research that used a single technique. Additionally, it is pioneering in applying machine learning methods to detect drift in the domain of loan processes.

The purpose of this paper is to detect different behavioral groups and the dominant patterns of customer shifts between segments of different values over time.

A new hybrid methodology is presented based on clustering techniques and mining top-k and distinguishing sequential rules. This methodology is implemented on the data of 14,772 subscribers of a mobile phone operator in Tehran, the capital of Iran. The main data include the call detail records and event detail records data that was acquired from the IT department of this operator.

Seven different behavioral groups of customer shifts were identified. These groups and the corresponding top-k rules represent the dominant patterns of customer behavior. The results also explain the relation of customer switching behavior and segment instability, which is an open problem.

The findings can be helpful to improve marketing strategies and decision making and for prediction purposes. The obtained rules are relatively easy to interpret and use; this can strengthen the practicality of results.

A new hybrid methodology is proposed that systematically extracts the dominant patterns of customer shifts. This paper also offers a new definition and framework for discovering distinguishing sequential rules. Comparing with Markov chain models, this study captures the customer switching behavior in different levels of value through interpretable sequential rules. This is the first study that uses sequential and distinguishing rules in this domain.