Characterization and analysis of errors in the value and cost overrun of public contracts in Colombia using machine learning algorithms

Authors

Felizzola, Heriberto

Devia, Daniela

Martin, July

Abstract

This paper presents the implementation of a machine learning algorithm for the detection and analysis of errors in the value and cost overrun in public contracts in Colombia, in the database of the Electronic Public Procurement System - SECOP I. The research begins with the characterization of the errors that were identified in the sample and which are made when entering the information into the system. Two algorithms were implemented, one to predict contracts with errors in value and the other to predict contracts with errors in cost overrun. The Random Forest classification algorithm is used, performances in accuracy of 0.91 and 0.76 are obtained, for the classification of errors in the value and cost overrun, respectively.