Large Language Models in a Local Environment for Generative AI-Based Automated Event Log Processing in Process Mining Techniques

Abstract

Process mining is vital for enhancing the operational efficiency and supporting strategic decision-making of organizations because it identifies bottlenecks and optimizes workflows. However, its analytical success depends fundamentally on the quality of event logs, and preparing these logs is a highly complex task. This complexity stems from diverse data storage formats across organizational systems, including varied log structures and inconsistent timestamps. Thus, analysts utilize public generative Artificial Intelligence (AI) services for data preparation, which demonstrates functional capabilities within usage limitations. However, the transmission of sensitive data to external servers in public generative AI systems poses data leakage risks. Moreover, manual intervention required in compiling the AI model-exported data is prone to errors and time consuming. Token limit violations in such models also cause service interruptions. To address these challenges, a local generative AI system was developed for automated event log transformation utilizing Ollama Llama 3.1. The system employed Python with the Streamlit framework, Pandas library, and Ollama integration. It thus implemented rule-based format detection algorithms, AI-assisted semantic column mapping, and a data transformation engine that managed complex conversions, including regular expression algorithms for timestamp standardization. This system was tested on real data obtained from a hospital in Thailand (128,917 events from 10,217 cases), during which it processed 4,194 events per second. It transformed complex data formats into standard event logs, which were then imported into a process mining software, Disco. Results indicated that this system addressed data security concerns by establishing automated processes, enabling organizations to leverage the benefits of AI in preparing data for process mining without data leakage risks.

Keywords: process mining, event log preprocessing, large language model, Ollama Framework, Artificial Intelligence (AI) local environment, automated data transformation

ดร.ภูริเดช อาภาสัตย์ และ ศ. ดร.วิเชียร เปรมชัยสวัสดิ์ – Dr. Poohridate Arpasat & Prof. Dr. Wichian Premchaiswadi. 2568 (2025). บทความ (Paper). บัณฑิตวิทยาลัย (Graduate School). ปรัชญาดุษฎีบัณฑิต สาขาวิชาเทคโนโลยีสารสนเทศ – Doctor of Philosophy Program in Information Technology. ปร.ด. เทคโนโลยีสารสนเทศ – Ph.D. in Information Technology. Bangkok: Siam University