Niklas Holtz¹ and Jorge Marx Gomez², ¹Future Research, Germany, ²Carl von Ossietzky Universitat, Germany

Risk classification plays a critical role in domains such as finance, insurance, and healthcare. However, identifying risks can be a challenging task when dealing with different types of data. In this paper, we present a novel approach using the Multimodal Transformer for risk classification, and we investigate the use of data augmentation for risk data through automated retrieval of news articles. We achieved this through keyword extraction based on the title and descriptions of risks and using various selection metrics. We evaluate our approach using a real-world dataset containing numerical, categorical, and textual data. Our results demonstrate that the use of the Multimodal Transformer for risk classification outperforms other models that only utilize textual data. We show that the inclusion of numerical and categorical data improves the performance of the model, particularly for risks that are difficult to classify based on textual data alone. Additionally, our research indicates that the utilization of data augmentation techniques yields enhanced performance outcomes in models. This methodology presents a promising avenue for enterprises to effectively mitigate risks and make well-informed decisions.

Risk classification, Multimodal Transformer, Data augmentation