The administrative procedural information system is currently in use in numerous cities and provinces throughout the nation. In order to enhance user experience and enable quicker and more easy access to public services for individuals, enterprises, and organizations, this system offers tools for registration for online public services via the Internet. However, according to current reported data in Soc Trang province, the rate of users registering online in the province is now relatively low. In order to help provincial authorities make informed decisions and implement the necessary policies to boost the number of people signing up to use online public services in the province, this study proposes a method to predict the amount of online records that will emerge in the near future. Data is collected from the province of Soc Trang’s administrative procedural settlement information system and preprocessed to fit the needs of the time series forecasting problem. We then employ deep learning models, including Transformer, LSTM, TiDE, and TCN, to predict the quantity of online profiles that will emerge in the near future. Additionally, we investigated the combining of the outcomes of the LSTM, Transformer, TCN, and TiDE forecasting models using the Ensemble Learning approach in an effort to increase the model’s accuracy rate. The experimental results show that the Ensemble Learning model achieves relatively high forecasting results, with the average RMSE, MAE, and R2-Score indexes being 206.07, 222.46, and 0.85, respectively.