<ul><li>Predictive Business Process Monitoring (PBPM) aims to forecast future outcomes of ongoing business processes, facing challenges like simultaneous events, class imbalance, and multi-level attributes.</li><li>To address these limitations, a suite of dynamic LSTM HyperModels has been proposed, integrating hierarchical encoding for event and sequence attributes, character-based decomposition of event labels, and pseudo-embedding techniques for durations and attribute correlations.</li><li>Specialized LSTM variants have been introduced for simultaneous event modeling, utilizing multidimensional embeddings and time-difference flag augmentation.</li><li>Experimental validation on four datasets shows significant accuracy improvements, with up to 100% accuracy on balanced datasets and F1 scores exceeding 86% on imbalanced ones. This approach enhances PBPM with more deployable models suitable for complex settings and contributes to improving temporal outcome prediction and data heterogeneity support.</li></ul>

Comprehensive Attribute Encoding and Dynamic LSTM HyperModels for Outcome Oriented Predictive Business Process Monitoring

Discover more