Association rule mining (ARM) is an important subject in data mining. Mining association rules is to find rules in the form of X → Y from the rule base that X and Y satisfy certain constraints. Class association rules (CARs) are a special type of association rules that apply to classification problems. Research in ARM and CARM (CAR mining) dates as far back as the early 1990s. Many algorithms have been proposed since then. However, all existing algorithms suffer the problem of inefficiency when working with datasets (rules) that are frequently updated, as any update requires recomputing the rules, subsequently taking a long time.

The authors propose an efficient algorithm to support incremental mining with record deletion. The algorithm uses a modified equivalence class rules-tree (MECR-tree) structure to store all frequent itemsets from the original dataset. The information in these nodes is updated as the classification algorithm proceeds. Efficient updates of the tree as the classification algorithm runs are key. A node in the MECR tree is updated using the information from deleted nodes only if the number of deleted nodes is less than a predefined threshold. Doing so reduces the amount of overall computation needed, that is, the total scans of the tree needed.

The authors conducted an extensive collection of performance evaluations. The results show that the proposed algorithm works very well compared to other existing algorithms when handling datasets with frequent updates.

Probably the result of trying to balance the amount of text, tables, and graphics on each page, readers often have to go back and forth in the paper to find the referenced tables or figures, which makes reading a bit more difficult than is necessary.

The algorithm presented in the paper is interesting. It has a practical impact on improving the performance of ARM involving updated nodes. Researchers in the area should find it useful.