A novel big data-based multicollinearity-eliminating feature extraction method for tourism demand forecasting

Integrating search engine big data into forecasting models has been proven to enhance prediction accuracy. However, the challenge of multicollinearity among search query variables limits the effective use of such data. To address this, this study proposes a multicollinearity-eliminating feature extraction (MEFE) method, integrated with a backpropagation neural network (BPNN), to create the MEFE-BPNN framework for tourism demand forecasting. Empirical results show that the proposed framework consistently outperforms traditional time series models and conventional feature extraction techniques in prediction accuracy. These findings highlight the MEFE-BPNN framework’s superiority and its potential as an advanced tool for big data-driven tourism demand forecasting.