Motion sickness (MS) remains a critical challenge in virtual reality (VR) applications, impacting user comfort and immersion. This study proposes a multimodal sensor fusion framework for detection and prediction of motion sickness in VR users. By integrating electrodermal activity (EDA), heart rate (HR), and facial skin temperature, the proposed framework employs effective temporal alignment, noise filtering, and adaptive normalization techniques to enhance data quality. A hybrid machine learning architecture combining gradient boosting decision trees and a transformer-based time-series model is developed to classify motion sickness severity levels and predict future symptoms. Experimental results demonstrate that the proposed method achieves high accuracy across multiple evaluation metrics (F1-score: 0.9855). Furthermore, the bimodal EDA+HR combination offers a lightweight alternative with minimal accuracy degradation, facilitating integration into consumer-grade VR systems. This research advances automated motion sickness detection and provides a pathway toward personalized VR experiences with enhanced comfort and usability.