In addition to the estimate of failure at the end of the year, the Predictor should be understood as an indicator of academic vulnerability. Insufficient performance in subjects, irregular attendance, accumulated gaps and contextual factors produce early signs that, if not addressed with appropriate interventions, tend to increase the risk of school failure.
Besides the variables directly associated with academic performance, such as grades and attendance, the results show the centrality of 6th grade as being a critical point in the school trajectory in the state network. This stage, marked by the transition of students from Elementary School to Junior High School, concentrates a higher incidence of failure.
In a complementary way, the age-grade distortion emerges as a factor associated with the increased risk of failure. Even when considering performance indicators throughout the school year, the presence of accumulated gaps maintains a significant influence on the probability of failure, showing that vulnerabilities built throughout the school path continue to impact results.
One of the main methodological differentials of the project is the structured incorporation of Explainable Artificial Intelligence (XAI) techniques. This approach enables both global analyses — identifying the most relevant factors associated with the risk of failure in the network — and individual analyses, allowing us to understand which specific elements increase or reduce the probability of failure of each student.
To enable the practical application of the results in educational management, estimates of the probability of failure were organized into five risk ranges: Critical, High, Upper Medium, Lower Medium and Low. The classification by levels allows structuring responses proportional to the identified vulnerability. Students in the critical range may require more intensive pedagogical interventions and systematic monitoring; intermediate levels can be directed to monitoring strategies and focused reinforcement; while those at low risk remain under regular preventive follow-up.
In addition to individual analysis, the model makes it possible to view and monitor information in aggregate form, by class, school, and municipality. This functionality expands the strategic reach of the tool, allowing identification of territorial patterns, support of regional planning, and guidance of the most efficient allocation of educational resources. In this way, the Failure Predictor is not limited to risk identification, but structures a complete flow of decision support, strengthening evidence-driven educational management.
As the next step of the work plan, expansion of the Failure Predictor is being planned for the High School segment of the state network. This expansion will make it possible to extend the logic of early identification of academic vulnerability to a phase marked by greater risks of dropout and abandonment.
The implementation of the Failure Predictor represents an advance in the institutionalization of the use of data science applied to educational management in Mato Grosso do Sul. More than estimating probabilities, the initiative strengthens the state's capacity to identify vulnerabilities, prioritize interventions, and promote more equitable educational trajectories. This is an important step in the consolidation of evidence-based public policies, with the potential of having a direct impact on the permanence, learning and future opportunities of students in the state network.
