Personality profiling, especially risk-taking propensity, is an important aspect in decision-making under conditions of uncertainty. As a measure of risk-taking behaviors, BART (Balloon Analog Risk Task) has been commonly applied by counting the number of balloon pumps before a given subject chooses to stop further risk-seeking behavior. Although useful for overall assessment, this approach may not reveal details of an individual’s risk profile. The application of machine learning in BART improves the assessment process as it breaks down each decision-making point to create a risk profile for every individual. This paper focuses on the re-estimation of decision thresholds in machine learning models for enhancing the reliability of risk propensity estimates by identifying how to adjust these thresholds at each decision-making point in terms of probabilities that enhance model’s risk behavior outcomes. This approach enriches the individual risk assessment and increases the efficiency of managerial psychometric assessments based on state-of-the-art machine learning algorithms. The study’s results show that model thresholding can alter conventional risk propensity evaluation tests into dynamic decision-making strategies for risk assessment, which provide important implications for both the behavioral sciences and the managerial decision-making practices.