Sedentary Behavior Classification for Wearable Sensors with a CNN-BiLSTM Model

2026-08-03

AI Summary

This study addresses the performance degradation in sedentary behavior recognition caused by sensor placement variability on the wrist. To mitigate this issue, the authors propose a transfer learning approach that adapts the CNN-BiLSTM model (CHAP), originally pretrained on hip-worn accelerometer data, to wrist-based data through zero-shot inference and fine-tuning strategies for sit/non-sit classification. This work presents the first empirical validation that models pretrained on hip-worn sensors can be effectively transferred to wrist-worn scenarios, highlighting the critical role of fine-tuning in enhancing classification accuracy. Experimental results demonstrate that the fine-tuned CHAP model significantly outperforms a Transformer model trained from scratch, thereby establishing the efficacy and superiority of the pretraining-plus-fine-tuning paradigm for cross-location sedentary behavior recognition.

Abstract

Accurate detection of sedentary behavior is important for studying health risks related to prolonged sitting, but posture-based classification remains challenging with wearable sensors, especially at the wrist. We study whether a deep learning model trained on hip-worn accelerometer data can transfer to wrist-worn accelerometer data for sitting versus non-sitting classification. We use CHAP, a CNN-BiLSTM model originally developed for hip accelerometers, and evaluate its zero-shot performance on wrist data as well as its adaptation through finetuning with varying amounts of labeled wrist data. Experiments are conducted on the iWatch dataset with ground-truth posture labels derived from wearable cameras. The hip-trained model performs strongly on hip data without retraining, but accuracy drops on wrist data due to sensor placement shift. Finetuning CHAP provides consistent advantages over transformer models trained from scratch. These findings suggest that hip-based pretraining provides a useful starting point for wrist deployment, while highlighting the need for wrist-specific adaptation to handle higher signal variability.

Problem

• sedentary behavior
• wearable sensors
• posture classification
• sensor placement
• accelerometer data

Innovation

• transfer learning
• CNN-BiLSTM
• sedentary behavior classification
• wearable sensors
• sensor placement adaptation