Connected Internet-of- Things (IoT) devices pose sev-eral privacy risks through the analysis of their encrypted network traffic. According to prior studies, packet size can be utilized to train a machine learning classifier for the identification of IoT de-vices because of their unique functionalities and traffic patterns. A recent defense technique aimed at addressing these privacy concerns efficiently is random segmentation [1]. This mechanism involves breaking down application messages into randomly sized chunks to obscure patterns in packet sizes. However, it leads to higher latency due to the increased number of packets and the additional packet header overhead. Furthermore, nonadaptive (or static) splitting in the original random segmentation approach is inappropriate for networks with dynamically changing conditions, which is common in smart homes. In this paper, we present an adaptive segmentation approach based on optimization, which adapts the splitting volume to changes in network usage. We formulate an optimization problem in order to maximize network traffic obfuscation while minimizing segmentation overhead. We evaluated our adaptive approach through simulations using real-world IoT data traces. Our results illustrate how the adaptive defense system adjusts its splitting parameters to enhance privacy protection, as measured by entropy, while minimizing the impact on transmission performance.