In this paper, we present a framework to deal with the fraud detection task with extremely few labeled frauds. We involve human intelligence in the loop in a labor-saving manner and introduce several ingenious designs to the model construction process. Namely, a rule mining module is introduced, and the learned rules will be refined with expert knowledge. The refined rules will be used to relabel the unlabeled samples and get the potential frauds. We further present a model to learn with the reliable frauds, the potential frauds, and the rest normal samples. Note that the label noise problem, class imbalance problem, and confirmation bias problem are all addressed with specific strategies when building the model. Experimental results are reported to demonstrate the effectiveness of the framework.
Identifying the fraud risk of applications on the web platform is a critical challenge with both requirements of effectiveness and interpretability. In these high-stakes web applications especially in financial scenarios, decision rules have been extensively used due to the rising requirements for explainable artificial intelligence (XAI). In this work, we develop a rule learning framework with rule mining and rule refining modules for addressing the learning efficiency and class imbalance issues while making the decision rules more broadly and simply applicable to risk management scenarios. On four benchmark data-sets and two large-scale data-sets, the classification performance, interpretability, and scalability of the framework have been proved, achieving at least a 26.2% relative improvement over the state-of-the-art (SOTA) models. The system is currently being used by hundreds of millions of users and dealing with an enormous number of transactions in Ant Group, which is one of the largest mobile payment platforms in the world.