Issue 1 (December)

Research Objectives: Build a dynamic imbalanced financial distress prediction model to solve concept drift I and imbalanced data distribution simultaneously. Research Methods: Based on the improved AdaBoost algorithm, ADA-CSSVM-TW model is built with cost sensitive support vector machine as the base classifier, and the data of Chinese manufacturing companies from 2010 to 2020 are used for the empirical analysis. Research Findings: The ADACSSVM-TW model I can significantly improve the prediction accuracy, with excellent performance and reliable robustness. Research Innovations: The cost sensitive support vector machine is used as the base classifier of the improved AdaBoost algorithm, and the dynamic imbalanced financial distress prediction model is built. Research Value: The research of this paper has important theoretical value and practical significance for Chinese listed companies to effectively prevent the occurrence of financial distress in practice.

With the development of Internet and E-commerce, picture advertising, as an important form of display advertising, has the characteristics of high visibility, strong readability and easy to obtain user recognition. An increasing number of Internet companies are paying attention to what kind of advertising pictures can attract more clicks. Recently, Deep Learning has been employed in automatic feature extraction, and has made remarkable achievements in the fields of computer vision, speech recognition, natural language processing and artificial intelligence. Compared with traditional shallow model, deep learning can automatically extract more complex features from simple features, which reduces the intervention of artificial feature engineering to a certain extent. Based on deep learning technology, this paper studies the prediction model of click through rate (CTR) for advertising, and proposes an end-to-end CTR prediction depth model for display advertising, which integrates the feature extraction of display advertising and CTR prediction to directly predict the probability of an advertisement image being clicked by users. This paper studies the deep-seated nonlinear characteristics through the multi-layer network structure of deep network, and carries out several groups of experiments on the private display advertising data set of a commercial advertising platform. The results show that the model proposed in this paper can effectively improve the prediction accuracy of CTR compared with other benchmark models.

Credit default identification technology usually requires a higher classification accuracy, and low rate of false positives, in view of the traditional credit default identification model based on machine learning, in dealing with high dimensional imbalance of financial transaction data on the problem of lower overall classification accuracy, along with the development of the large data, the credit transaction data of large-scale and high dimension feature, It is necessary to ensure the efficient and timely construction of the credit default recognition model. Semi-supervised learning can reduce the cost of manual labeling, and train the model in time by making full use of a large amount of unlabeled data and a small amount of labeled data, which has important research significance in the field of credit default recognition. At the same time, the traditional supervised learning model can only recognize the default behaviors that have happened before, but the default behaviors are variable, and the semi-supervised learning algorithm can recognize the unknown default behaviors. From the perspective of semi-supervised learning, DBN, as a semi-supervised deep learning framework, utilizes its feature learning ability and iForest unsupervised learning algorithm's ability to identify abnormal behaviors to propose a default identification method by use of the DBN-Iforest. This paper first introduces the experimental data, including the data source and the pretreatment process of the original data. Then the performance evaluation criteria applied to all experimental models in this paper are introduced in detail. Then the steps of DBN-iForest algorithm are described in detail. Next, algorithms for iForest optimization are introduced, including particle swarm optimization algorithm and simulated annealing optimization algorithm. Finally, through comparative analysis of experimental results at different levels, the advantages of the DBN-Iforest model are proved, the classification performance is improved, and the difficulty of data resource shortage in credit default recognition is overcome. The semi-supervised default identification method by use of the DBN-IForest fully improves the classification performance of DBN and iForest by using less label data.

In order to solve the location routing problem (LRP) by means of  energy saving and environmental protection, we propose a LRP optimization model with bi-objectives of minimizing the carbon emissions and the distribution costs. To address the shortcomings of traditional heuristics in solving large-scale LRPs with poor generality and low efficiency, an improved fast non-dominated ranking genetic algorithm (NSGA-II) is proposed and applied to the LRP optimization. In order to improve the convergence and optimization ability of the algorithm, an improved method of adaptive crossover operator and adaptive population size adjustment is introduced on the basis of the original method. With the benchmark test example solved, the algorithm is able to design an accurate, efficient and intelligent scheduling scheme for solving the established LRP model. Compared with the traditional heuristic algorithm in terms of the overall quality of the solution and the convergence efficiency of a single solution, the feasibility and effectiveness of the proposed method are verified.

P2P (Peer to Peer) online lending is an emerging Internet finance mode that gathers small-amount fund lending to fund demander. This paper draws on the existing credit risk assessment research, combines rationality, science and other principles, according to the characteristics of the famous online loan platform, collects borrower information and combines computer technology to design a borrower credit risk assessment system. In this paper, we make improvements based on the famous LightGBM algorithm (Light Gradient Boosting Machine). Firstly, In the process of data input, the improved Convolutional Neural Network CNN model is adopted to extract features from the data. Specifically, the Global Average Pooling(GAP) layer is adopted to replace the full connection layer to improve the Convolutional Neural Network. This paper first proposes a P2P online loan default prediction model based on GCNN-LightgBM. The model integrates the advantages of the improved Convolutional Neural Network and LightGBM model, and realizes the efficient prediction of network loan default. Then, in order to improve the accuracy of P2P online loan borrower default prediction, this paper proposes a new model based on LightGBM and Bagging (LGB-BAG). LGB-BAG uses LightGBM as the base learner. With the help of LightGBM, which can effectively reduce the deviation of the model, and Bagging, which can reduce the variance of the model, the volatility of the prediction is further reduced (F1 variance), so that the LGB-BAG model has smaller deviation and variance, and the prediction effect is further improved. In our ablation experiment, the proposed model (GCNN-LGB-BAG) obtained an AUC of 0.86 and an accuracy of 0.97, both of which outperformed all benchmark models. This paper uses actual data to identify the loan risks of P2P online lending platforms, aiming to provide investment reference for investors and methodological support for relevant online lending regulators.