Handling data absence caused by communication failures in distributed multi-sensor systems is a critical challenge in wearable technology, where computational efficiency and energy consumption are key constraints. This paper addresses the impact of data absence caused by communication failures in multi-sensor wearable systems on the classification of body postures using LSTM neural networks, by implementing and evaluating three extrapolation methods: Zero-Order Holder (ZOH), First-Order Holder (FOH), and Second-Order Holder (SOH). Results show that FOH and SOH outperform ZOH, particularly when transitioning between states, with SOH achieving the best performance in moderate noise scenarios. While ZOH is less effective in prolonged data absence, its simplicity and low cost make it viable for resource-constrained systems. These methods significantly mitigate the impact of missing data, improving model robustness and reliability. This study provides a foundation for enhancing wearable system’s performance in real-time applications and paves the way for integrating advanced sensing strategies and cardiorespiratory indicators in future research.