Supervising Hotspot Sharing(HS) networks remains a persistent and suspended challenge, attracting widespread attention from academia and industry. With the popularity of encrypted traffic, traditional methods relying on plaintext fields in message headers have almost become ineffective. As the contradiction between the increasing ineffectiveness of traditional methods and the demand for more refined technologies becomes increasingly prominent, Huawei posted the technical challenges and requirements related to HS supervision on Huang Danian Chaspark in 2023. Compared to the binary classification problem of HSed-behaviors identifying accurately, detecting the number of devices in an HSed-network based on passive measurement is more challenging. To aim at this challenge, by leveraging the unique traffic patterns of HSed-devices, we propose a novel measure, Packet Interval Statistics (PIS), that captures the interval variations in network packets. Based on PIS, we design a set of features and develop the PIS-EnsemNet model, which integrates ensemble learning and deep learning techniques. Experimental results demonstrate that PIS-EnsemNet significantly outperforms state-of-the-art regressors regarding MAE, MSE, RMSE, and \(R^2\) scores. Our method converges rapidly during training and exhibits excellent generalization capability, validated by its high F1-Score of 97.09% in device count prediction. This innovative approach offers a promising solution for supervising HSed-networks accurately.