Top-K High Utility Episode Mining from a Complex Event Sequence

Utility episode mining has emerged as an interesting and challenging research topic in data mining. It finds applications in anomaly detection, biomedical data analysis, predicting stock trends etc. The number of high-utility episodes that can be extracted from a sequence depends upon the value of minimum utility threshold. It is often difficult for a user to find a suitable threshold value which fits their purpose. The sequence can generate many high-utility episodes at low threshold value and very few episodes at higher threshold values. In order to relieve the user from this tedious task, we propose an algorithm for mining top-k high utility episodes from a complex event sequence. The parameter k can be set by the user according to his/her needs. We also propose effective strategies for raising the threshold value in order to prune the search space effectively. We conduct extensive experiments on real and synthetic datasets and the experimental results demonstrate the effectiveness of our proposed strategies in terms of total execution time and the number of candidate episodes generated.