Abstract
The Brazilian National School Feeding Program (PNAE, acronym in Portuguese) establishes the obligation that at least 30% of federal resources transferred to the executing boards be allocated to the acquisition of food from family farms. However, the results achieved with the program reveal high regional heterogeneity. This study used the machine learning approach (particularly classification trees) to identify factors that most influence compliance with PNAE legislation in the Northeast and South regions of Brazil based on data from 2,650 municipalities referring to the year 2017. The analysis of these two regions is justified by the high presence of family farming, although with significant differences in the fulfillment of the 30% target: 77% of municipalities in the South meet this target, compared to 38% in the Northeast region. The results of the survey indicate that differences between the two regions are associated with factors such as financial limitations of the municipalities and less formalization of agricultural establishments. Moreover, aspects such as the size of the municipality, degree of centralization in the acquisition of food, and profile of local agricultural production also influence compliance with legislation. Therefore, public policies should consider regional specificities to improve adherence and the effectiveness of implementation on the regional level.