Using machine learning to discover indicators of well-being

Regardless of their personal, professional and social circumstances, individuals can experience different levels of satisfaction, fulfillment and happiness in life. This general measure of life satisfaction, commonly referred to as “well-being”, has been the subject of numerous psychological studies.

A better understanding of the many factors that contribute to well-being could help design personalized and targeted interventions aimed at improving individuals’ levels of flourishing. While many studies have attempted to delineate these factors, few have done so by exploiting the advanced machine learning models available today.

Machine learning models are designed to analyze large amounts of data, reveal hidden patterns, and make accurate predictions. Using these tools to analyze data collected in previous studies in neuroscience and psychology could help shed light on the environmental and genetic factors that influence well-being.

Researchers from the Vrije Universiteit Amsterdam and the University Medical Center Amsterdam recently set out to study the predictors of well-being by analyzing a large dataset collected in the Netherlands over a ten-year period using machine learning. Their findings, published in Nature Mental Healthidentifies several factors that could predict the well-being of the general population.

“Effective personalized well-being interventions require the ability to predict who will thrive and who will not, and an understanding of the underlying mechanisms,” Dirk H. M. Pelt, Philippe C. Habets, and colleagues wrote in their paper.

“Using longitudinal data from a large population cohort (the Dutch Twin Registry, collected from 1991 to 2022), we aim to build machine learning prediction models for adult well-being from the exposome and genome, and identify the most predictive factors (N between 702 and 5874).”

Pelt, Habets, and their colleagues used machine learning models to analyze the Dutch Twin Registry, a dataset collected from a large population cohort over an 11-year period. The data were collected from the same children when they were approximately 3, 5, 7, 10, 12, 14, and 15 years old, as well as from three separate waves of adult participants.

The Dutch Twin Registry dataset includes genetic information called polygenic scores (i.e. genome), as well as information about the participants’ environment (i.e. general exposome) and psychosocial conditions (i.e. specific exposome). The researchers trained three different machine learning models, dubbed XGBoost (XGB), SVM, and RF, on this large amount of data.

They then used a powerful technique called Shapley Additive Explanation (SHAP) to explore the contributions of different characteristics to the predictions made by the three models. Their analysis revealed that the model’s predictions of well-being were based on various environmental and psychosocial factors related to participants’ satisfaction, happiness, and quality of life.

“The specific exposome was captured by parent and self-reports of psychosocial factors from childhood to adulthood, the genome was described by polygenic scores, and the general exposome was captured by the linkage of participants’ zip codes to objective registry-based exposures,” Pelt, Habets, and colleagues wrote.

“It’s not the genome (R²= −0.007 (−0.026–0.010)), but the general exposome (R²= 0.047 (0.015–0.076)) and in particular the specific exposome (R²= 0.702 (0.637–0.753)) were predictive of well-being in a set of independent tests. Adding the genome (P = 0.334) and general exposome (P = 0.695) independently or jointly (P = 0.029) beyond the specific exposome did not improve prediction.

Overall, the researchers observed that participants’ genetic predispositions (i.e., their genome) did not predict their reported well-being, while environmental and psychosocial factors did. The factors they found to be most predictive of well-being were optimism, personality traits, social support, neighborhood dynamics, and housing characteristics.

“Our results highlight the importance of longitudinal monitoring and the promise of different data modalities for predicting well-being,” the researchers wrote.

This recent study by Pelt, Habets, and colleagues suggests that specific environmental, social, and psychological circumstances contribute most to individuals’ subjective well-being. Moving forward, the findings could inform the development of personalized interventions aimed at improving life satisfaction for specific individuals, while potentially inspiring the further use of machine learning to explore factors influencing well-being.

© 2024 Science X Network