The Beast-GB model combines machine learning and behavioral science to predict people's decisions

The key objective of behavioral science research is to better understand how people make decisions in situations where results are unknown or uncertain.

The ability to predict people's choices in these situations is extremely advantageous as it helps them draft effective initiatives aimed at encouraging people to make better decisions for themselves and others in their community.

Technology researchers (Israel Institute of Technology) and various US laboratories have recently developed a new computational model called Beast-GB, which has been found to predict people's decisions in situations with risk and uncertainty.

The proposed model is outlined in a paper published in Natural human behaviorcombines advanced machine learning algorithms with behavioral science theory.

“While human decision research is rich in competing theories, none of the whole context will certainly accurately predict human choices,” Ori Plonsky, the first author of The Paper, told Tech Xplore.

“We organized the CPC18, a 'choice prediction competition', to see which ideas actually work. This “choice forecasting competition” allowed anyone to submit a computational model to predict people's decisions under risk and uncertainty. I was particularly interested in knowing whether data-driven machine learning, theory-driven behavioral models, or hybrids embedding behavioral theory within ML would expand.

A new machine learning model developed by Plonsky and his colleagues draws from a behavioral science framework known as Beast (Best Taimatiate and Sampling Tools). This is a model based on psychological theory, which we previously found to accurately predict people's decisions.

“We assume that Beasts will choose under risk and uncertainty to confuse several strategies, such as minimizing the likelihood of immediate regret or hedging for the worst outcome,” explained Plonsky.

“Each strategy has been translated into 'behavioral functions'. This is a concise formula that captures how sensitive decision makers should be to that consideration in a particular choice task. In addition to functionality based on these theories, we then fed purely objective task descriptors to extreme gradient boosts (machine learning algorithms known to be extremely useful in predictive tournaments).

With enhancements implemented by researchers, the Beast-GB model can analyze behavioral data and derive the decisions driving motivation, and the impact of these motivations in a variety of decision-making scenarios.

Notably, Beast-GB won the CPC18 Choice Prediction Competition in 2018, winning 93% of predictable variation in FED data and 96% in follow-up tests using a 40x larger data set.

“Beast-GB was better than dozens of mainstream behavioral models and purely data-driven machine learning,” Plonsky said.

“Only 2% of the training data we broke deep neural networks trained on all training data. The model accurately predicts the choices people make in new experiments that imply capture common human choice patterns. Finally, we use it to improve and enhance our capabilities as it is used to improve and enhance fundamental interpretable theory of behavior, not only predicting human decisions.”

This recent study highlights the promise of machine learning models drawn from behavioral science to predict people's decisions and responses in real-world scenarios. In the future, Beast-GB and other similar models could guide the design of new, large-scale interventions aimed at improving people's decisions through Nudges, incentives, or other behavioral science-based strategies.

Plonsky and his colleagues will ultimately work with policymakers and other stakeholders involved in the design or implementation of behavioral science initiatives. This allows them to test the model “wild” and test its possibilities in a real-world setting while also providing insights that inform further advancements.

“Other recent publications suggest that human decision-making and other behaviors can be predicted very effectively using sophisticated data-driven machine learning methods, such as large-scale language models coordinated with large-scale behavioral data,” added Plonsky.

“We will now continue to explore when and how beast-like theories can enhance such data-driven methods when predicting behavior. Specifically, we plan to expand the realm of research by including natural language decision problems.”

©2025 Science X Network