In credit underwriting modelling, BOMB can help in identifying risky customers and at the same time screen more revolvers (customers who pay interest regularly for revolving) than transactors, thus achieving twin goals of higher revenue and improved profitability.
BOMB can combine two very different modelling objectives into one single machine learning exercise achieving both the objectives efficiently. Naturally, if both the objectives are dealt individually, significant costs, in terms of labour and time, are incurred to achieve the objectives. When two different machine learning models are deployed, reaching the intersection solution of both the problems requires high degree of manual intervention, which if not done carefully, can lead to pure illogical results.
What is BOMB?
Our machine learning algorithm referred as Bayesian Optimization for Multiple Boosting (BOMB) solves real world analytical problems while also addressing the constraints that a typical modeler is faced with when applying Machine Learning optimization algorithms separately. At its most basic, machine learning uses algorithms that receive and analyze input data to predict output values within an acceptable range. The data driven approach of machine learning algorithm makes them popular in any modeler’s tool kit.
Generally, all other hyperparameter optimization techniques work well when the objective function is a single metric (like R square, MAPE, Gini, KS, etc.), but serving dual or multiple objective function optimization requirement is typically hard to fulfill. Our Bayesian Optimization System (BOS) is a scientific hyperparameter tuning algorithm using Gaussian process (GP) for exploitation & exploration tradeoff in its learning algorithm.