Machine Learning Models for Predicting the Performance of Risk Management Applications running in Cloud

Machine Learning (ML) is assuming an increasingly predominant role in the context of predicting application performance. Its significance lies in its ability to provide a black-box approach, which does not require a deep understanding of the application internals. This approach enables accurate predictions without delving into the details of the system, which may either be unavailable (as in the cloud) or require the formulation of highly complex models. In our study, we leveraged ML to forecast the execution time of an industrial cloud-based application dealing with risk measures as part of an Internal Model to implement Solvency II regulations for insurance companies, providing users with valuable insights into its duration based on ML algorithms. By conducting a comparative analysis of multiple models, it was determined that XGBoost is the most effective choice. The results indicated robust forecasting accuracy for intermediate times, albeit showing limitations for shorter and significantly longer durations, primarily attributed to data scarcity in those specific settings. The study achieved an overall forecasting Mean Absolute Percentage Error of 14%, highlighting the potential of ML in predicting the performance of complex industrial applications.