Wi-Fi Resources Forecasting by Using Multivariate Prediction Based on Input Data Clustering Using an Ensemble of Decomposed Transformer Models

The exponential growth of connected devices and the increasing dependence on wireless networks necessitate efficient Wi-Fi resource management. Accurate forecasting of Wi-Fi resources, including predicting future user access numbers, is crucial for optimizing advertising strategies and ensuring targeted user engagement. This paper introduces a novel approach to Wi-Fi resource prediction utilizing an ensemble of decomposed transformer models combined with input data clustering. Our methodology leverages the power of transformer models to handle multivariate time series data, enhancing the accuracy and reliability of predictions. By clustering input data, we improve the model’s ability to generalize across different usage patterns and scenarios. Experimental results demonstrate that our approach significantly outperforms traditional prediction models, providing robust and precise forecasts of Wi-Fi resource demand. This advancement holds the potential for better network performance management and more effective advertisement targeting in dynamically changing environments.