Machine Learning in Procurement with a View to Equity

The application of machine learning to big data from tenders published in Italy provides significant benefits to public administrations and economic operators, including improved procurement processes. Quantitative results from our study show a 96.5% accuracy using XGBoost models for predicting the presence of tender variations during contract execution. These models estimate the likelihood of variations in upcoming tenders: their correct prediction is a valuable tool because variations avoidance reduces completion time and costs of public contracts. Additionally, explainable AI tools help the description graphically and intuitively of the analyzed data. They also allow the analyst to highlight potential biases in tender participation and their awards, contributing to fairer public procurement. The results of their application to public tenders show that strong differences in the Italian country exist with a consequent lack of equity. Finally, the application of recommendation systems on the tender notices shows they are an effective cognitive tool to search for similar tenders and retrieve the actors involved, such as public administrations or economic operators. The precision score of the answers is above the value of 90% for the 74.15% of the queries. The chapter describes the tasks that permit the achievement of the above objectives.