Class-weighted neural network for monotonic imbalanced classification

In real life scenarios, classification problems with the characters of monotonicity constraints and imbalanced class distribution widely exist. However, at present, the research on this kind of problem is still rare. Traditional algorithms designed only for monotonic classification and imbalanced classification are not available for monotonic imbalanced classification. So far, there is only one approach specially designed for monotonic imbalanced classification problems, which is based on the resampling technique. In this paper, from the algorithmic point of view, we propose a weighted single-hidden-layer feedforward neural network (WMCS-SLFN) based on multi-objective genetic algorithm, where both the monotonicity constraints and the imbalanced distribution are considered. Additionally, in order to improve the generalization capability of WMCS-SLFN, we put forward a selective ensemble strategy for WMCS-SLFN based on the 0–1 knapsack problem, which can generate an ensemble of WMCS-SLFN with the optimal prediction accuracy under the monotonicity constraints. Contrast experiments conducted on eight monotonic imbalanced datasets verify the effectiveness of our proposed methods, and moreover, the experimental results analyzed by Wilcoxon statistical test highlight the advantage of our work significantly.