In the process of developing interventions targeting health behaviors (e.g., smoking cessation, safe sex, healthy nutrition), intervention planners interact with multiple types of information coming from various sources (e.g. theory, previous research, input from stakeholders). Intervention planners then need to meaningfully integrate this information into an analytical framework that guides the construction of the intervention components. Commonly used frameworks often oversimplify human behaviors despite acknowledging their inherent complexity. For instance, frameworks include chains of lists (in logic models) instead of richer structures considering loops and the dynamicity of relations between determinants. There is thus a lack of systematic approaches to effectively form such rich structures in health behavior interventions. While integrating many aspects of complexity in logic models may be possible, it also runs the risk of producing artifacts that become equally complex instead of equipping intervention planners and health researchers with decision-making tools to estimate the effects of interventions. Consequently, effective intervention development requires (i) techniques of systematic integration of information of different types and from various sources, which can account for (ii) the complex structure and functioning of interactions between the identified behavioral determinants, and (iii) will support the testing of intervention scenarios. In this paper, we show how a hybrid approach of Fuzzy Cognitive Maps and machine learning may support various aspects of complexity in intervention design by meeting all three requirements.