Aims: School non-completion is a public health and educational concern in most countries. This study sought to identify the strongest predictors of the non-completion of upper secondary education based on register data. Methods: A cross-validated elastic net regression analysis was used to predict school non-completion in a population of 2696 students in the city of Jyväskylä, Finland. The register data included data from the primary social and healthcare register and the educational register. Results: The non-completion rate was 13.1% (13.4% for males, 12.8% for females). The non-completion of upper secondary education was best predicted by the following seven features (ordered from strongest to weakest): unauthorized absences (odds ratio (OR) = 2.27), out-of-home placement (OR = 2.23), average grade when leaving lower secondary education (OR = 0.73), an anxiety/depression diagnosis (OR = 1.43), visits to child guidance and family counselling centres (OR = 1.17), family poverty (OR = 1.11) and the grade point average in the 5th Grade (OR = 0.95). Conclusions: Register data can be utilized to find the strongest predictors of school non-completion. Predictors support multidisciplinary actions preventing non-completion by providing both early signals to target actions more specifically and indicators for monitoring the impact of preventative actions.