To deal with the systematic risk of financial institutions and the rapid increasing of loan applications, it is becoming extremely important to automatically predict the default probability of a loan. However, this task is non-trivial due to the insufficient default samples, hard decision boundaries and numerous heterogeneous features. To the best of our knowledge, existing related researches fail in handling these three difficulties simultaneously. In this paper, we propose a weakly supervised loan default prediction model WEAKLOAN that systematically solves all these challenges based on deep metric learning. WEAKLOAN is composed of three key modules which are used for encoding loan features, learning evaluation metrics and calculating default risk scores. By doing so, WEAKLOAN can not only extract the features of a loan itself, but also model the hidden relationships in loan pairs. Extensive experiments on real-life datasets show that WEAKLOAN significantly outperforms all compared baselines even though the default loans for training are limited.

Endogenous volatile organic compounds (VOCs) can reflect human health status and be used for clinical diagnosis and health monitoring. Dimethylamine and ammonia are the signature VOC gases of nephropathy. In order to find a potential gas sensitivity material for the detection of both signature VOC gases of nephropathy, this paper investigated the adsorption properties of dimethylamine and ammonia on Al-and Ga-doped BN monolayers based on density functional theory. Through analyzing the adsorption energy, adsorption distance, charge transfer, density of states, and HOMO/LUMO, the results indicated that the adsorption effect of Al-and Ga-doped BN monolayers to dimethylamine and ammonia is probably good, and these nanomaterials have the potential to be applied for nephropathy monitoring and clinical diagnosis.

No abstract

Based on multi-core learning, this paper uses the method of skin color segmentation, fingertip distance detection and extraction and gesture contour extraction, which realizes the fusion of local features and overall features and makes the accuracy of hand gesture classification, reach 92%. Compared with traditional gesture recognition based on binocular camera, our algorithm using multi-core learning has the same accuracy as the recognition based on binocular camera, and the complexity and cost of our algorithm are lower. Thus, our algorithm has broader application prospects. Compared with gesture recognition based on monocular camera, our algorithm is more suitable for the background environment of gesture classification. In other words, it has better generalization ability.

Customer segmentation is critical for auto insurance companies to gain competitive advantage by mining useful customer related information. While some efforts have been made for customer segmentation to support auto insurance decision making, their customer segmentation results tend to be affected by the characteristics of the algorithm used and lack multiple validation from multiple algorithms. To this end, we propose an auto insurance business analytics approach that segments customers by using three mixed-type data clustering algorithms including k-prototypes, improved k-prototypes and similarity-based agglomerative clustering. The customer segmentation results of these algorithms can complement and reinforce each other and demonstrate as much information as possible to support decision-making. To confirm its practical value, the proposed approach extracts seven rules for an auto insurance company that may support the company to make customer related decisions and develop insurance products.