Advancing Idiographic Methodology for Data-Driven Treatment Personalization: Comparing ARIMAX, Network, and Tree-Based Approaches

Idiographic approaches offer a promising path to personalized psychological interventions, though guidance on selecting optimal methods remains limited. Researchers have applied various time series, network, and tree-based methods to identify individual-level process-outcome relationships, yet their comparative effectiveness is under-evaluated. We examined six advanced methods: two ARIMAX-based time series models (bivariate and multivariate i-ARIMAX), two network approaches (GIMME, indSEM), and two tree-based algorithms (Boruta and our novel Time Series Boruta, or tsBoruta). Simulations tested these methods across diverse longitudinal data conditions, varying in linear/nonlinear effects, autoregressive patterns, trends, interactions, predictor counts, and sample sizes. We evaluated performance using sensitivity, specificity, precision, and F1 scores. For linear effects, i-ARIMAX, iBoruta, and tsBoruta achieved the highest F1 scores and specificity. Tree-based approaches performed best at detecting nonlinear effects and interactions. Methods that failed to account for time-series dependencies showed significant performance drops as these factors increased. In an empirical application, i-ARIMAX and tsBoruta replicated the group-level loneliness–depressed mood link while revealing substantial individual-level heterogeneity and distinct predictor patterns, informing potential personalized targets. Combining i-ARIMAX and tsBoruta offers a practical workflow that balances sensitivity and specificity, captures both linear and nonlinear processes, and supports data-driven, idiographic psychological treatment planning.