@techreport{oai:ipsj.ixsq.nii.ac.jp:00239562,
 author = {Elsen, Ronando and Sozo, Inoue and Elsen, Ronando and Sozo, Inoue},
 issue = {19},
 month = {Sep},
 note = {In this paper, we enhance the ability of Large Language Models (LLMs) to interpret accelerometer data for detecting physical fatigue. While sensors are increasingly used to monitor physical conditions, analyzing their data remains challenging. We focus on improving LLMs' question-answering capabilities, specifically for data generated from physical activities. To address the complexities of sensor data, we introduce and compare two input formats―list-based and graph-based representations. Using models like gpt-3.5-turbo and gpt-4o-mini in both zero-shot and few-shot scenarios, we find that graph-based formats significantly boost LLM performance, achieving a top F1-score of 67.50%. This research highlights the importance of refining data formats and enhancing model capabilities to improve LLMs' effectiveness in health monitoring applications., In this paper, we enhance the ability of Large Language Models (LLMs) to interpret accelerometer data for detecting physical fatigue. While sensors are increasingly used to monitor physical conditions, analyzing their data remains challenging. We focus on improving LLMs' question-answering capabilities, specifically for data generated from physical activities. To address the complexities of sensor data, we introduce and compare two input formats―list-based and graph-based representations. Using models like gpt-3.5-turbo and gpt-4o-mini in both zero-shot and few-shot scenarios, we find that graph-based formats significantly boost LLM performance, achieving a top F1-score of 67.50%. This research highlights the importance of refining data formats and enhancing model capabilities to improve LLMs' effectiveness in health monitoring applications.},
 title = {Leveraging Large Language Models to Enhance Understanding of Accelerometer Data on Physical Fatigue Detection Question Answering},
 year = {2024}
}