FERNANDES, B. A.; http://lattes.cnpq.br/7579312038291904; FERNANDES, Bruno Andrade.
Abstract:
In face of the need to manage public capital efficiently and transparently, and the significant and growing allocation of resources in the state of Paraíba towards public works, this paper suggests a facilitator for public administration in the state and its constituent municipalities. Aimed at aiding decisions regarding the allocation of public capital and enhancing the efficiency of expenditure on public works, this study proposes training a predictive model for government project failures. By leveraging investigation of the "open data from SAGRES - TCE/PB," machine learning models were trained using Extreme Gradient Boosting (XGBoost) with different subsets of features, both with unbalanced and balanced data, capable of performing binary classification between project success or failure. Additionally, an improvement in model accuracy was observed when training with the aggregation of certain characteristics.