A clustering-based flexible weighting method in AdaBoost and its application to transaction fraud detection

AdaBoost is a famous ensemble learning method and has achieved successful applications in many fields. The existing studies illustrate that AdaBoost easily suffers from noisy points, resulting in a decline of classification performance. The main reason is that it increases the weights of all misclassified samples (especially noisy points) in the same way so that the influence of noisy points can hardly be weakened. In this paper, the clustering algorithm is used to dynamically decide noisy points in the process of iterations. More precisely, we compute a misclassification degree for every cluster in every iteration that is used to decide if a misclassified sample is a noisy point or not in the current iteration. Furthermore, we propose a flexible method to update the weights of the misclassified samples. The experimental results on 22 public datasets show that our method achieves better results than the state-of-the-art methods including AdaBoost, AdaCoast, LogitBoost, and SPLBoost. We also apply our method to the transactions fraud detection, and the experiments on our real big dataset of transactions also illustrate its good performance.