Past research has made substantial efforts to understand the radicalization process. However, to date, quantitative research in this area is undeveloped. Therefore, we used a machine learning approach to identify predictive variables in the context of Islamist radicalization contrasting radicalizing individuals with a meaningful control group. Our algorithm i) predicted the radical status with an accuracy as high as 80.36% and ii) uncovered the most predictive features of radical Islamists among the explored variables. A tendency to withdraw, being male, and a high lack of perspective were highly indicative of being radical, while not feeling excluded, the mosque or friends being the source of radicalization, and low authoritarianism were features highly indicative of not being radical. Contributing in a new way to terrorism research, our study might add to the understanding of radicalization and inspire the work of practitioners to improve detection and prevention efforts.