AI analysis reveals top predictors of first grade success: Insights from multifactorial screening students’ early days of school

Early prediction of students' primary school academic performance is crucial for planning timely interventions for those expected to struggle. Random Forest, Gradient Boosting, and Extra Trees AI techniques were used to study 2549 first graders to predict end-of-year academic performance and identify critical predictors for targeted interventions. The Extra Trees model used EPIS (Empresários pela Inclusão Social) screening to yield high F1 scores above 99.7 %, accurately identifying passing and failing students. Out of 142 variables, the top ten strongest predictors of academic success were categorized into three areas: 1) teacher perceptions of cognitive function, particularly attention (distractibility) and psychomotor slowing; 2) early literacy and numeracy skills, notably vowel identification, verbal fluency, and seriation; and 3) parental education level, which was found to be more predictive than socioeconomic indicators. These results demonstrate the effectiveness of AI models like Extra Tree in predicting end-of-year academic outcomes, from early screenings to managing at-risk students. However, the exceptionally high F1 score, despite precautions against overfitting, should be cautiously approached due to potential biases affecting generalizability. Findings also suggest that comprehensive training in executive functions and early literacy and numeracy skills in preschool and early primary education might enhance academic outcomes, even for students from disadvantaged backgrounds. Involving caregivers in promoting academic values further supports these efforts. Educational stakeholders are urged to prioritise early initiatives, such as primary school Head Start programs and a strong preschool curriculum, to reinforce foundational skills, enrich students' academic journeys, and foster the social mobility that education seeks to achieve.