Forecasting intergroup contest outcome with inter-brain neural synchronization

821 

Abstract Submission 

Xiaochun Han¹, Jiaxin Yang², Yina Ma^2,3

¹Faculty of Psychology, Beijing Normal University, Beijing, China, ²State Key Laboratory of Cognitive Neuroscience and Learning, Beijing Normal University, Beijing, China, ³Chinese Institute for Brain Research, Beijing, China

Xiaochun Han  
Faculty of Psychology, Beijing Normal University
Beijing, China

Jiaxin Yang  
State Key Laboratory of Cognitive Neuroscience and Learning, Beijing Normal University
Beijing, China

Yina Ma  
State Key Laboratory of Cognitive Neuroscience and Learning, Beijing Normal University|Chinese Institute for Brain Research
Beijing, China|Beijing, China

Forecasting the outcome of intergroup contests is of great interest in human society, from sports betting to warfare victor prediction. However, accurately predicting the outcome of intergroup contests remains challenging, even for experts (Bunker & Susnjak, 2022). Therefore, the development of an accurate predictive model for intergroup contests holds both theoretical and practical significance. By leveraging the strengths of simultaneous neural recording of individuals engaged in real-time intergroup contests (Yang et al., 2020) and advanced machine-learning techniques (Kohoutová et al., 2020), the current study aims to investigate whether and if so, how a neural predictive model based on intra- and/or inter-brain activity can effectively forecast the outcome of intergroup contests.

We adopted a previously established intergroup contest game paradigm in the laboratory consisting of three attackers and three defenders and simultaneously recorded the brain activities of six participants using functional near-infrared spectroscopy (fNIRS) (Figure 1A&B). The fNIRS channels were positioned over the right temporal-parietal junction and the dorsolateral prefrontal cortex (Yang et al., 2020; Zhang et al., 2023). To forecast the contest outcome, we extracted the activities of the two brain regions during the decision-making phase of the contests, including two sets of intra-brain neural activities, i.e., activation level and functional connectivity, and two sets of inter-brain neural synchronization, i.e., neural synchronization between individuals within the group and between groups (Figure 1C). We developed the predictive neural model in the Discovery cohort (N = 30 sessions, with 6 participants in each session) using a support vector machine algorithm with a radial basis kernel (Figure 1D, Patle et al., 2013). Then, we examined the key brain features in forecasting contest outcomes. Lastly, we examined the neural model's generalizability in an independent Testing cohort (N = 36 sessions).

The predictive model developed in the Discovery cohort using both intra- and inter-brain neural activities achieved an accuracy of 76.18% (p < 0.0002; Figure 2A&B). We further compared the contributions of intra- and inter-brain features in predicting contest outcomes. Interestingly, only inter-brain neural synchronizations made significant contributions to prediction accuracy, indicated by substantial decreases in prediction accuracy when shuffling these feature sets (decreased acc = 9.92% and 12.17%, pFWE = 0.02 and 0.01, respectively for within- and between-group neural synchronization). Moreover, we showed that the model utilizing only inter-brain neural synchronizations achieved an increased predictive accuracy of 92.12% (pFWE < 0.0004; Figure 2C&D). However, when relying solely on intra-brain neural activities, the model exhibited limited ability to forecast the contest winner (acc = 58.23%, pFWE = 0.068; Figure 2E&F). Consequently, we opted to employ the predictive model exclusively based on inter-brain neural synchronization for contest outcome forecasting, which we termed the Predictive Neuromarker of Intergroup Contest Outcome. The feature importance analysis for each of the neuromarker's features showed that the inter-brain neural synchronizations of the attacker group played a predominant role in predicting the outcome of intergroup contests. Lastly, we applied the neural model to an independent Testing cohort and revealed a high accuracy of 81.94% (p < 0.0002) in the generalization test.

The current work has developed a neural predictive model that incorporates both within- and between-group inter-brain neural synchronization, enabling accurate prediction of the winner of intergroup contests. These findings shed light on the development of sophisticated neural models for complex social interactions by considering the dynamic interaction of brain activities between different individuals.

Social Interaction ¹

Decision Making

Classification and Predictive Modeling ²

NIRS

Computational Neuroscience

Machine Learning

Modeling

Multivariate

Near Infra-Red Spectroscopy (NIRS)

Social Interactions

Other - Intergroup conflict