In the Netherlands, up to 23% of adolescents are affected by anxiety disorders and for younger children estimates range from 5-20% (Verhulst, van der Ende, Ferdinand, & Kasius, 1997; Bögels, 2007). Childhood anxiety is associated with a host of future problems, such as substance abuse, academic failure, and even suicidal behaviour (Woodward & Fergusson, 2001). The impact is enormous, both on a personal level and in terms of the associated societal costs (Greenberg, et al., 1999; RIVM, 2013). Clearly, effective prevention and treatment programs are urgently mandated but despite evidence-based interventions being readily available, their outcomes are often disappointing and variable (Reynolds, Wilson, Austin, & Hooper, 2012). Existing one-size-fits-all interventions are at best moderately effective long-term, and most importantly, not tailored to differences in individual responsiveness (James, James, Cowdrey, Soler, & Choke, 2013; Cartwright, Hatton, Roberts, Chitsabesan, Fothergill, & Harrington, 2004). This project’s aim is to contribute to a new approach to the prevention and treatment of anxiety in children by developing new methodology to uncover and tailor interventions to individual response patterns, using a unique biofeedback virtual reality game (DEEP). DEEP targets key developmental factors associated with anxiety and consists of an immersive virtual underwater world, which players can explore freely, controlling their movements through breathing. We aim to uncover individual profiles of behavioural change and tailor DEEP to these profiles in real-time, combining dynamical systems theory and machine learning for automated classification of individual in-game behaviour.