Tourism demand forecasting: a novel multi-channel imaging model

Purpose This study aims to introduce an innovative multi-channel imaging technique aimed at mitigating deep learning overfitting and facilitating the automatic extraction of features from limited 1D data. Design/methodology/approach The proposed framework consists of five key component: dimensionality reduction, sequence image generation, image stitching, feature extraction and model training. It converts 1D multi-temporal data into multiple 2D images utilizing Markov transition field, Gramian angular field and recurrence plot. These single-channel images are stitched into a larger n-channel image, which is processed by a convolutional neural network for feature extraction and forecasted using a long short-term memory network. Findings The results demonstrate that the proposed multi-channel imaging technique outperforms all benchmark models. This conversion captures underlying patterns and enhances information transmission. Additionally, models with multi-time series configurations perform better than single-time series setups, highlighting that data are more crucial than advanced models in forecasting. Originality/value This pioneering study explores the role of non-economic variables in tourism forecasting. The proposed multi-channel time series imaging model not only applies to tourism but also offers potential for interdisciplinary applications.