From Logs to Language and Vision: A Review on Integrating Multimodal Data into Predictive Process Monitoring

Abstract

Predictive Process Monitoring (PPM) forecasts future outcomes of business processes based on event logs. However, current approaches often overlook rich multimodal data – such as text, audio, video, and sensor inputs – that are increasingly generated in real-world settings. This paper presents a systematic literature review of 89 studies on multimodal data integration in PPM and related BPM tasks. We identify dominant modalities (e.g., sensor and text), methodological trends, and application domains, along with key challenges such as data fusion complexity, scarcity of labeled datasets, and model interpretability. At the same time, we highlight opportunities, including context-aware prediction, synthetic data augmentation, and real-time decision support. We propose future research directions which lay the foundation for advancing multimodal, intelligent BPM systems in complex environments.