Promoting Simple Agents: Ensemble Methods for Event-Log Prediction

2026-04-23 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

This research compares lightweight automata-based n-gram models against neural architectures like LSTMs and Transformers for next-activity prediction in streaming event logs. Experiments conducted on synthetic patterns and five real-world process mining datasets reveal that n-grams, when configured with suitable context windows, achieve accuracy comparable to neural models while demanding significantly fewer computational resources. Unlike the unstable performance observed in windowed neural architectures, n-grams demonstrate stable and consistent accuracy. The study also introduces a novel ensemble method, the "promotion algorithm," which dynamically selects between two active models during inference. This approach reduces the memory consumption and latency typically associated with classical ensemble methods like voting, while matching or exceeding the accuracy of non-windowed neural models on real-world datasets at a lower computational cost.

Key takeaway

For AI Engineers and Research Scientists developing event-log prediction systems, consider integrating n-gram models with the proposed "promotion algorithm" ensemble. This approach offers a compelling balance of high accuracy and significantly reduced computational overhead compared to complex neural architectures, making it ideal for resource-constrained environments or applications requiring stable, consistent performance. Evaluate its fit for your specific streaming data needs.

Key insights

N-gram models offer stable, resource-efficient next-activity prediction comparable to neural networks.

Principles

• N-grams provide stable, consistent accuracy.
• Ensemble methods can improve n-gram performance.

Method

The "promotion algorithm" dynamically selects between two active models during inference to reduce ensemble overhead.

In practice

• Use n-grams for resource-constrained prediction.
• Consider dynamic ensembles for efficiency.

Topics

• Event-Log Prediction
• Next-Activity Prediction
• N-grams
• Neural Architectures
• Ensemble Methods

Best for: AI Engineer, Research Scientist, AI Scientist, Machine Learning Engineer

Related on AIssential

• Promoting Simple Agents: Ensemble Methods for Event-Log Prediction — N-gram models offer competitive accuracy for event-log prediction with significantly fewer resources than neural networks.
• Grammar of the Wave: Towards Explainable Multivariate Time Series Event Detection via Neuro-Symbolic VLM Agents — Neuro-symbolic VLM agents utilize Event Logic Trees for explainable, zero-shot time series event detection from natural language descriptions.
• N-Grams and Markov Assumptions: The First Predictive Models of Language — N-grams were the first predictive language models, using statistical counts over short word sequences to anticipate the next word.
• Predicting Overnight Stock Changes Post-Earnings: Tomorrow’s Price Hidden in Today’s Language — Earnings call language, processed with domain-specific NLP, predicts overnight stock movements.
• Explainable AI based ensemble model for the identification of Schizophrenia prodromal phase — An XAI-integrated ensemble ML model accurately predicts Schizophrenia prodromal symptoms for early clinical intervention.
• Ensembles of Ensembles of Ensembles: A Guide to Stacking — Ensembling diverse models across multiple layers enhances predictive performance and robustness for tabular and time series data.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.