Evaluating the performance of personality-based profiling in predicting physical activity.

Area of Science:

• Psychology
• Statistics
• Data Science

Background:

• Clustering individuals based on personality traits is common in marketing, medicine, and social sciences.
• This approach simplifies data and aids decision-making but involves information loss through variable discretization.
• The impact of this information loss on predicting external outcomes remains under-investigated.

Purpose of the Study:

• To assess the accuracy of predicting physical activity using a clustering approach.
• To compare the predictive accuracy of clustering versus a dimensional approach using continuous variables.

Main Methods:

• Utilized survey data from 20,573 individuals.
• Assessed prediction accuracy of physical activity using both clustering and dimensional approaches.
• Included Big-Five personality traits and physical activity data.

Main Results:

• A four-cluster solution, based on standard criteria, achieved only 60-70% prediction accuracy compared to the dimensional approach.
• The dimensional approach, using raw continuous variables, demonstrated higher predictive power.

Conclusions:

• Conventional statistical criteria for determining clusters may not yield optimal solutions for predicting external outcomes.
• Information loss during clustering can significantly impact the accuracy of predicting variables like physical activity.