Personality trait prediction by machine learning using physiological data and driving behavior

Abstract

This article explores the influence of personality on physiological data while driving in reaction to near crashes and risky situations using Machine Learning (ML). The objective is to improve the driving assistance systems in considering drivers’ characteristics. Physiological and behavioral data were recorded in sixty-three healthy volunteers during risky urban situations and analyzed using 5 ML algorithms to discriminate the driver’s personality according to Big Five Inventory and STAI trait. Seven step process was performed including data pre-processing, Electrodermal Activity (EDA) time windows selection (one by one backward and forward approach comparison with a pseudo-wrapped), personality traits assessment, input algorithms parameters optimization, algorithm comparison and personality trait cluster prediction. ROC Area Under the Curve (AUC) was used to describe improvement. The pseudo-wrapped/all possibilities method comparison resulted in 8.3% on average for all personality traits and all algorithms (% of ROC AUC of backward and forward approach). The ROC AUC for the detection of the personality ranged between 0.968 to 0.974 with better detection of Openness, Agreeability and Neuroticism. Use of association between Neuroticism, Extraversion and Conscientiousness previously defined in the literature slightly improve personality detection (maximum ROC AUC of 0.961 to 0.993 for cluster). Results are discussed in terms of contribution to driving aids. This study is one of the first to use machine learning techniques to detect personality traits using behavioral and physiological measures in a driving context. Additionally, it questions input parameters optimization approach, time windows selection, as well as clustering and association of personality trait for detection improvement.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 44 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.