Abstract:
Ukrainian households found themselves forced to adopt negative mechanisms for coping in order to endure the hardships of war in Ukraine. These strategies, including reducing necessary expenses or selling assets, may have long-term effects on household’s capacity to effectively handle money and maintain good health. This thesis explores the use of advanced machine learning (ML) algorithms to identify the primary predictors of the harmful coping strategies that Ukrainian households used throughout the conflict. Furthermore, it investigates the effect of several variables, such as proximity to frontline, financial and economic status, and household characteristics, on the likelihood of utilizing negative coping techniques. After finding the optimal set of features for each outcome, Lasso model was used to assess magnitude of the association found in the complex imbalanced dataset between the predictors and the outcomes. Paired with cost-sensitive learning, we were able to identify the significant predictors of negative coping mechanisms. The associations uncovered between the predictors and outcomes revealed two distinct early warning patterns for policymakers: those that serve as protective factors and those that serve as risk factors. These insights, in turn, serve as early warnings for policy makers to target vulnerable households, helping to prevent resorting to negative coping mechanisms.
