<p>Internet of Healthcare Things (IoHT) extends the Internet of Things (IoT) by integrating smart technologies into healthcare systems. A key enabling technology is Human Activity Recognition (HAR), which utilizes sensor-based data to automatically identify and interpret human behaviors, supporting intelligent, context-aware healthcare applications. This paper proposes a sensor-based HAR framework centered on a novel feature selection (FS) algorithm derived from an enhanced version of the Ivy Algorithm (IVYA), a nature-inspired metaheuristic that models the growth mechanisms of ivy plants. Although IVYA successfully addresses a range of challenging optimization problems, it still suffers from significant limitations, including premature convergence, loss of population diversity, imbalanced exploration-exploitation, and limited adaptability in complex search spaces. These shortcomings hinder its ability to effectively solve high-dimensional and multimodal optimization problems, often leading to suboptimal solutions. Therefore, an improved version, termed DAGE-IVY, is proposed to tackle these limitations. DAGE-IVY incorporates four enhancement strategies: (1) Dynamic Drift Search (DDS) to balance exploration and exploitation, (2) Adaptive Diversity Preservation Strategy (ADPS) to maintain population diversity, (3) Search-Hide (SH) strategy for adaptive exploration control, and (4) Adaptive Growth (AG) rate to align the search process with the optimization dynamics. The binary variant of DAGE-IVY is utilized in the proposed HAR framework to select the most relevant features, using classification error (evaluated by a K-Nearest Neighbors classifier) as the fitness metric. Experiments are conducted using two publicly available HAR datasets to assess the effectiveness of the proposed framework. Comparative analysis against other optimization algorithms revealed that DAGE-IVY achieved the best overall performance.</p>