This study introduces the tortoise attention algorithm (TAA), a novel computational tool for accurately measuring children’s concentration during learning activities. The algorithm integrates weighted behavior duration and temporal stability metrics to calculate a comprehensive concentration score. We conducted three studies to evaluate the effectiveness of the algorithm. Study 1 demonstrated that TAA’s concentration scores significantly predicted performance on math tasks but found no significant relationship with executive function performance, as measured by the ANT test. Study 2 revealed that TAA outperformed humans in predicting children’s math performance, underscoring the algorithm’s ability to mitigate biases inherent in human assessments of fidgeting behaviors. Study 3 further showed that while human evaluations of concentration were consistent across classroom and café settings, they failed to align with actual learning outcomes. These findings highlight that TAA provides an objective and reliable tool for evaluating concentration, enabling educators to refine teaching strategies and improve learning outcomes.