Efficient Selection of Globally Optimal Rules on Large Imbalanced Data Based on Rule Coverage Relationship Analysis

Abstract

Rule-based anomaly and fraud detection systems often suffer from massive false alerts against a huge number of enterprise transactions. A crucial and challenging problem is to effectively select a globally optimal rule set which can capture very rare anomalies dispersed in large-scale background transactions. The existing rule selection methods which suffer significantly from complex rule interactions and overlapping in large imbalanced data, often lead to very high false positive rate. In this paper, we analyze the interactions and relationships between rules and their coverage on transactions, and propose a novel metric, Max Coverage Gain. Max Coverage Gain selects the optimal rule set by evaluating the contribution of each rule in terms of overall performance to cut out those locally significant but globally redundant rules, without any negative impact on the recall. An effective algorithm, MCGminer, is then designed with a series of built-in mechanisms and pruning strategies to handle complex rule interactions and reduce computational complexity towards identifying the globally optimal rule set. Substantial experiments on 13 UCI data sets and a real time online banking transactional database demonstrate that MCGminer achieves significant improvement on both accuracy, scalability, stability and efficiency on large imbalanced data compared to several state-of-the-art rule selection techniques.

Coupling learning of complex interactions

“…In [12], several new types of patterns were identified, including pair patterns, cluster patterns, and approaches to quantify combined patterns such as combined association rules [71,72]. In [32], the interactions between rules are analyzed to avoid the duplication caused by multiple rules with overlapping functions while maintaining the detection quality of rule-based risk management. (6) Testing coupled data.…”

“…In fact, real-world data sets are all embedded with more or less couplings. In our practice, we have tested couplings in real-world business, including capital markets [10,8,54], online banking business [32], recommender systems including social media [33], and web text [17]. We have shown that even a very commonly used data set, or the highly manipulated UCI data, can be used for coupling learning evaluation, as we showed in [59,60,61] for coupled clustering and coupled ensemble clustering, although the couplings may not be very strong, and the performance difference by incorporating couplings may not be very obvious.…”

“…Couplings between patterns form the basis of the need for pattern relation analysis [5], which may be explored from semantic, structural and knowledge representation perspectives. In [32], relations between multiple rules are explored to identify those rules which optimally contribute to the most cost-effective detection of online banking risk.…”

Fraud Detection in Payments Transactions: Overview of Existing Approaches and Usage for Instant Payments

“…Li et al [26] proposed a solution for the selection of globally optimal (business) rules for detecting fraud. The proposed MCGminer algorithm is based on the Max Coverage Gain metric, which scores how good a rule performs globally.…”

Non-IIDness Learning in Behavioral and Social Data

“…These techniques are general and can be widely used and expanded for analyzing complex behavioral and social problems. Interested readers can find a detailed introduction from the cited references and also find some of our other efforts on quantifying similarity in categorical [37] and numerical objects [41], coupled clustering by incorporating coupled object similarity [37], analyzing non-IIDness at the method level to explore couplings between clusterings for coupled ensemble clustering [42] and considering the relations between patterns [28] and rules [43] for pattern relation analysis.…”