EURO-Online login
- New to EURO? Create an account
- I forgot my username and/or my password.
- Help with cookies
(important for IE8 users)
2433. FICO Decision Optimizer – Generating predictive models with Action Effect
Invited abstract in session WD-63: Complexity and Financial Patterns, stream OR in Banking, Finance and Insurance: New Tools for Risk Management.
Wednesday, 14:30-16:00Room: S14 (building: 101)
Authors (first author is the speaker)
| 1. | Claudio Gambella
|
| FICO House, FICO, Xpress Optimization | |
| 2. | Livio Bertacco
|
| FICO Optimization | |
| 3. | Sebastien Lannez
|
| Xpress Optimization, FICO | |
| 4. | Ben Willcocks
|
| FICO House, FICO, Xpress Optimization | |
| 5. | Brendan del Favero
|
| FICO, Xpress Optimization, US | |
| 6. | Ryan Weber
|
| FICO, Xpress Optimization, US |
Abstract
The FICO Decision Optimizer (DO) application is an optimization software package to perform optimal assignment of treatments to a portfolio of customers. DO leverages optimization algorithms with the goal to empower non-OR professionals with a tool that creates and solves Generalized Assignment Problems, without requiring them to formulate the models. DO considers different kinds of constraints (e.g., budget, ratio) and allows users to generate highly interpretable decision trees.
The DO models depend on structural inputs (e.g., portfolio or customer attributes) and uncertain inputs (predictable target values, such as the impact of treatments on customers). While the DO interface enables configuring and editing of the predictive models required to produce the uncertain data, developing Predictive Causal Models usually requires additional tools and knowledge for a Business Analyst. Furthermore, there is inherent historical data bias, because the historical actions are likely to be targeted on certain segments. Consequently, there may be significant data gaps for some actions.
To simplify the user experience, we developed an Action Effect (AE) approach to predict the impact of the treatment assignment on every customer of the portfolio. We present the logic behind the quadratic programming models that create the predictive models. We conclude our presentation with a discussion about the advantages of this methodology over standard unconstrained linear regression.
Keywords
- Financial Modelling
- Practice of OR
- Analytics and Data Science
Status: accepted
Back to the list of papers