Toda City's AI Non-Attendance Risk Score — A Saitama Truancy Early-Warning Pilot With Built-In Human Safeguards

Toda City Board of Education, in Saitama Prefecture, Japan, ran a pilot under Japan's Children and Families Agency 'Children Data Collaboration Demonstration Project' across 18 public elementary and junior-high schools (roughly 12,000 students), building an AI risk-scoring model with Uchida Yoko, PKSHA Technology and its subsidiary Argonaut.

The pilot aimed to combine attendance and tardiness records, health-room visit logs, academic achievement survey results, school-life questionnaires and counselling records into a non-attendance ('futoko') risk score per student, restricted to principals and administrators, to enable earlier outreach to at-risk students.

The city published guidelines banning discriminatory use of the risk scores, offered families an opt-out, and restricted data access to school leadership; teachers used the risk flags to prioritise students for direct outreach and support.

The model flagged 1,193 of roughly 12,000 students as elevated-risk; teachers used the flags to prioritise 265 of them for direct outreach. In a post-pilot survey, 90% of participating schools rated the predictions as highly reliable.

An independent education economist (Keio University's Makiko Nakamuro) publicly cautioned against over-reliance on the algorithm, stressing it should support rather than replace human judgement; the pilot ran for one school year across 18 schools with no confirmed citywide or national rollout yet documented.