The development of new mathematical techniques, the improvement of computer processing power and the increased availability of possible explanatory variables has changed the way companies model their risks. This has caused the financial industry to move towards the use of new machine learning methods, such as neural networks, and away from older methods such as generalised linear models, to assist in the decision-making process. i.e. whom to grant loans to.
The goal of this talk is to show how biplot methods can be used to visualise the various input factors and the output of the machine learning black box. This will assist in quantifying and understanding the black box machine learning model by visualising:
- How the model’s decision probabilities vary by different variables along the biplot axes and therefore identify the most significant variables influencing the model.
- The difference in the prediction probabilities and therefore the model decision made by linear and non-linear models for different combinations of input values.
- The distance for an individual case to the decision boundary and which variable needs the smallest amount of change to impact the decision made for the case.
This will help identify the individual inputs that has the most significant impact on the decision made for the case. Because the results are all visualised in two dimensions, they are easy to explain and intuitive to understand. This makes them suitable for decision makers to quickly review the results and justify the decisions made based on the output of the black box model.