Contracting in Brazilian public administration: A machine learning approach

Abstract

AbstractThe risk of non‐fulfilment of a contract can harm public administration or even interrupt public services. Therefore, models that assist manager decision making in the audit and control of contracts with a higher disqualification risk may be important tools, with economic and even social repercussions. In this article, public contracts are classified with respect to the risk of non‐compliance with their terms of delivery. The quantitative tools used are statistical and machine learning models, similar to credit risk rating of loans. As dependent variables, the models use data found in electronic databases present in e‐government implementations. A previously classified listing of suspended companies is used as a proxy for risky contracts, as it contains private companies which failed with their contractual obligations. The classification techniques utilized are logistic regression, k‐nearest neighbours, discriminant analysis, support vector machine and random forests. Although the methods can be applied to any government with electronic procurement and contracts systems, Brazilian data is used to illustrate the benefits of contract governance for emerging economies. It is concluded that the credit rating techniques used directly apply to contractual risk in public administration. Considering real public administration contract data, the classification algorithm that generates the best performance is k‐nearest neighbours.