Idiographic Prediction of Behaviors and Experiences Using Symbolic Classification

Psychologists are increasingly arguing that persons are best conceived as complex, dynamical, and nonlinear systems. Due to heterogeneity in individual experiences and behaviors, and the heterogeneity in the processes which lead to these behaviors, an idiographic approach may be necessary to fully understand the relevant factors driving individual behavior. Prior studies have demonstrated that idiographic machine learning models can predict future behaviors and experiences from intensive measurements of personality, affect, and situations (Beck & Jackson, 2022; Soyster et al., 2022). However, an important goal in psychology is not only to predict behaviors but to understand the relevant processes that drive them. Machine learning approaches, such as LASSO or elastic net regression, while interpretable, may fail to capture nonlinear dynamics or complex interactions, limiting both their predictive power and inferential utility. Conversely, techniques such as random forest, support vector machines, and neural networks are capable of attaining high performance on complex, high dimensional datasets, but can suffer from a lack of interpretability. Thus, there is a critical need for an approach capable of (1) capturing complex dynamics (2) while remaining interpretable and (3) performing well on relatively short time series. Symbolic regression, which uses an evolutionary search procedure to discover concise mathematical equations to capture functional relationships among variables, is a promising approach (c.f. Makke & Chowla, 2024). In this project, we aim to implement idiographic symbolic classification models to forecast procrastination, loneliness, and social interaction across three samples.