Methodology of Using Empirical Distributions of Binary Prediction Scores to Solve Business Optimization Problems  
Author Andrei V. Shcheprov

 
Co-Author(s) Srinivas Krovvidy; Hernando A. Vera

 
Abstract Abstract - Many business questions can be formulated as constraint optimization problems. The paper considers one general class of business problems which are coupled with underlying binary classification algorithms. In modern practical applications, such binary classifiers generate prediction scores and the question of selecting optimal thresholds is of great importance. The paper presents a methodology that addresses this problemusing empirical distributions of prediction scores. We call this approach an Empirical Compartmental Optimization. The paper also shows how classification model hyper-parameters can be tuned based on the formulated business requirements. We run several computer simulated experiments to illustrate some properties of the proposed method and to analyze its computational stability.

 
Keywords machine learning, binary classification, constraint optimization, empirical distribution, bisection method
   
    Article #:  DSIS19-54
 
Proceedings of ISSAT International Conference on Data Science & Intelligent Systems
August 1-3, 2019 - Las Vegas, NV, U.S.A.