How Intel technology stack improved model training in Advancing Analytics' custom solution for the financial sector

The ability of AI to transform and improve processes for businesses is seen across many industries and has influenced the development of new technologies to support these algorithms and improve their performance. Virtualised cloud machines have added agility and adaptability to software stacks across the globe, allowing businesses to quickly adopt and deploy a greater range of hardware to keep up with market demands and modernisation. 

Cloud computing is being adopted at all scales in every industry, but the propensity for adoption of the latest processor released varies due to business requirements. Advancing Analytics partnered with Intel to make use of their OneAPI AI Analytics toolkit to improve the performance of a model which was being trained to detect and prevent potentially fraudulent home insurance claims. 

 Artificial Intelligence can be crucial to combating insurance fraud in several ways: 

• Pattern Recognition and Anomaly Detection: Machine learning algorithms excel at identifying patterns and anomalies in large datasets. They can detect patterns and anomalies that indicate fraudulent claims such as patterns of overstated claims, multiple claims from the same individual or group, or claims that deviate significantly from a customer's profile. 

• Predictive Analytics: AI can learn from historical data to predict the likelihood of fraud. By examining past claims that were fraudulent and identifying the characteristics of past fraudulent claims, AI can flag suspicious claims that match these criteria. 

• Customer Risk Profiling: By analysing a customer's behaviour, transaction history, and other personal data, AI can create risk profiles that help insurers assess the likelihood of fraudulent claims. 

 Advancing Analytics had a project from a client in the insurance industry, where they were tasked with reducing training run times on a custom-built model by integrating hyperparameter tuning into the process with the Scikit-learn optimisation library from the Intel OneAPI AI Analytics toolkit. The training was done on the end client’s virtual machines, which were hosted on an Azure Databricks Standard_d8_v3 cluster, where several generations of Intel machines are available. Advancing Analytics carried out benchmarking on 1st generation Intel® Xeon® 8171M 2.1GHz (Skylake) machines and achieved a significant training run time reduction of 29%, greatly improving the performance of the model. This was achieved on a training set of 60043 rows and 69 features.