Credit card fraud (CCF) is a significant threat to cardholders and financial institutions. CCF detection against this threat is challenging due to extremely imbalanced data (EID). EID involves extremely few instances of fraud for training and an extremely high risk of overlooking fraud. While class balancing or oversampling techniques can address the former problem by punishing negative classes or augmenting the positive data, they do not mitigate the latter. In contrast, the cost-sensitive learning approach targets only the high risk of false negative errors. Therefore, existing approaches are insufficient to solve all the issues of the EID problem. Based on the LightGBM (Light Gradient Boosting Machine) framework, this study introduces two novel machine-learning methods: the class balancing cost-harmonization LightGBM (CB-CHL-LightGBM) and the oversampling cost-harmonization LightGBM (OS-CHL-LightGBM). The new approaches combine class balancing or oversampling technology with LightGBM to solve the EID problem comprehensively. They enhance the efficacy of LightGBM in CCF detection scenarios. Experimental results on three CCF datasets indicate that the two proposed methods outperform LightGBM in several crucial performance metrics. For example, compared with the original LightGBM, CB-CHL-LightGBM or OS-CHL-LightGBM can increase the F2-score from 0.77 to 0.83 for the first dataset, from 0.77 to 0.86 for the second dataset, and from 0.70 to 0.82 for the third dataset. However, adding class balancing, oversampling, and cost-harmonization loss separately to LightGBM may not obtain better results.

Stock market prediction (SMP) is a challenging task due to its uncertainty, nonlinearity, and volatility. Machine learning models, such as artificial neural networks (ANNs) and support vector regression (SVR), have been widely used for stock market prediction and achieved high performance in the sense of "minimum errors." In the context of SMP, however, it is more meaningful to measure the performance using "minimum cost." For example, a false positive error (FPE) could result in a big trading loss, while a false negative error (FNE) might just miss a chance. For a "cautious" investor, fewer FPEs are preferable. In fact, cost-sensitive learning has been used in areas such as fraud detection and medical diagnosis. In our earlier study, we proposed a false-sensitive method called focal-loss LightBGM (FL-LightGBM) for SMP by introducing a cost-aware loss in LightGBM, which is known to be a fast and efficient gradient-boosting learning algorithm for solving large-scale problems. FL-LightBGM, however, still assumes that all false negative errors (or false positive errors) contribute equally to the final cost. Such learned trading strategies might be useful only for an investor who is always "aggressive" or "cautious." In practice, some errors may result in irreversible loss, so it is important to measure the cost based on "data" rather than the investor’s character. In this paper, we propose a new method called cost-harmonization loss-based LightGBM (CHL-LightGBM), in which the cost for each datum can be calculated dynamically based on the difficulty of the datum. To verify the effectiveness of CHL-LightGBM, comparisons have been made among LightGBM, XGBoost, decision trees, FL-LightGBM, and CHL-LightGBM for stock predictions on data from Shanghai, Hong Kong, and NASDAQ Stock Exchanges. The simulation results show that although there is no significant difference between CHL-LightGBM and other models on the accuracy and winning rate, CHL-LightGBM obtained the highest annual return on all the test data.

Stock market prediction (SMP) is challenging due to its uncertainty, nonlinearity, and volatility. Machine learning models such as recurrent neural networks (RNNs) have been widely used in SMP and have achieved high performance in terms of “minimum error”. However, in the context of SMP, using “least cost” to measure performance makes more sense. False Positive Errors (FPE) can lead to significant trading losses, while False Negative Errors (FNE) can result in missed opportunities. Minimizing FPE is critical for investors. In practice, some errors may result in irreparable losses, so measuring costs based on data is important. In this research, we propose a new method called generalized loss CNN-BiLSTM (GL-CNN-BiLSTM), where the cost of each datum can be dynamically calculated based on the difficulty of the data. We verify the effectiveness of GL-CNN-BiLSTM on Shanghai, Hong Kong, and NASDAQ stock exchange data. Experimental results show that although there is no significant difference in the accuracy and winning rate between GL-CNN-BiLSTM and other methods, GL-CNN-BiLSTM achieves the highest rate of return on the test data.

In this paper, we propose a multi-branch neural network architecture named Coupled Modular Neural Network (CMNN). A CMNN is a network consisting of $\beta $ closely coupled sub-networks, where $\beta $ is termed as the branching factor in this paper. We call the whole network a super-graph and each sub-network a sub-graph. Each sub-graph is a stand-alone neural network and shares a common block with other sub-graphs. To effectively leverage the super-graph we propose a simple but easy-to-implement Round-Robin-based learning algorithm. Each training iteration contains two phases. In the first phase, we choose a sub-graph in a Round-Robin fashion and train it using knowledge of the super-graph (distillation). In the second phase, we fine-tune the super-graph based on the updated sub-graphs. This algorithm produces a different copy of the super-graph at each iteration which acts as an improved teacher network for the sub-graph; and a different copy of one of the sub-graphs which functions as a new building block for the super-graph. To validate and test CMNN and the proposed algorithm, we conduct experiments on CIFAR-10, CIFAR-100, Tiny ImageNet and a private On-Road-Risk (ORR) datasets. Empirical results on all these four datasets indicate that we not only obtain a strong sub-graph network, the learning framework can also produce strong ensemble performance which substantiates the diversity introduced throughout the learning framework.