Work zone intrusion prediction is vital for enhancing the efficacy of road work zone safety systems and preventing intrusion accidents. However, false alarms have been identified as one of the major gaps limiting a wider application of such safety systems in road work zones. Existing deterministic approaches to predicting intrusions fail to account for the variability of human driving behaviours and vehicle trajectories, resulting in false alarms. This paper established several critical driving scenarios of vehicles approaching road work zones in Virtual Reality (VR). Driving behaviour and vehicle trajectory data were collected from 19 participants. A linear mixed-effects (LME) model was developed to statistically analyse the effects of driving scenarios and in-vehicle intrusion alerts on drivers’ reaction time (RT) and last-minute avoidance manoeuvres. The results revealed that more critical scenarios elicited both shorter reaction time and more aggressive avoidance manoeuvres, while in-vehicle intrusion alerts significantly affected RT only. The outcomes of this study contribute to a better understanding of the variability in driving behaviours and vehicle trajectories, facilitating risk assessment of work zone intrusions in future research.