In our 1st blog, we introduced you to our fully adaptive and data-driven customer behaviour detection system, FUSION, that by construction is 100% Covid-proof, and drastically improves the AML detection process.
The main purpose of FUSION is to provide the user with a fully data-driven methodology to construct segments of customers who exhibit similar (financial) behaviour.
The clusters are derived recursively thanks to a powerful Unsupervised Learning algorithm that scores each customer according to how “normal” their behaviour is within their segment.
Given its intuitive explanatory power, versatility and computational capacity, FUSION is a SaaS that can be used across different industries and for different purposes.
FUSION’s algorithm was designed following the principles below:
The users of FUSION will be able to upload their dataset onto the platform. Based on our experience, a typical dataset is made up of:
In case of raw data, we are able to assist you and engineer the features for you.
The key step is to fit a cluster to an Isolation Forest model, which essentially provides each customer with a similarity score that is a representation of the average number (across the trees) of iterations necessary to isolate a customer from others.
Intuitively, as you can see from the charts above, the higher the score, the more iterations required and, hence, the most similar to other customers in the cluster a customer is.
Once the normality score is assigned, the algorithm computes the SHAP values for each customer: these values determine the marginal impact of each feature on every customer’s similarity score, helping to understand the structure of the cluster. Using these marginal values for ranking the feature importance within each cluster, the algorithm continues to sub-cluster each segment until it reaches a homogenous status.
We will get back to the algorithm functioning in a more detailed blog post in the coming future.
In the next blog, we will first deep dive into FUSION’s applications, in particular, the Financial Services and Retail sectors.
Get in touch ASAP if you want to book your demo of FUSION.