Research on Fall Pre-Warning Algorithm Based on ISSA-DBN

To achieve efficient and rapid identification and prediction of imbalance behaviors in a fall pre-warning system, we propose an imbalance behavior identification method based on an improved Sparrow Search Algorithm (SSA) optimized Deep Belief Network (DBN). The Random Forest algorithm is employed to evaluate the importance of data features by calculating the Gini index, effectively reducing computational complexity and mitigating the impact of low-contribution data on the identification results. An adaptive weighted Golden Sine Method is introduced to enhance the SSA, which is then used for parameter optimization of the DBN, including determining the network structure. The optimized DBN enables fast and accurate identification of imbalance behaviors. Experimental results demonstrate that the prediction accuracy for all three categoriesimbalance behavior, falls, and activities of daily living (ADL)exceeds 99 %. Additionally, the system achieves a recognition accuracy of over 98 % for each category of imbalance behavior, ensuring real-time fall pre-warning capability.