Enhanced backpropagation neural network accuracy through an improved genetic algorithm for tourist flow prediction in an ecological village

Extant tourism studies on predicting tourist flow often adopt Backpropagation Neural Network (BP-NN) and Genetic Algorithm-Backpropagation Neural Network (GABP-NN). However, those models cannot well address the challenge of nonlinear complexity of tourists’ mobility, and fuzzy decision-making due to abrupt urgencies and foul weather. The current study proposes “Adaptive Multi-population Genetic Algorithm Backpropagation (AMGA-BP)”, which features a novel double-layer ladder-structured chromosome design for simultaneous optimization of network structure and weights. Experimental results demonstrate the AMGA-BP model achieves superior performance with a Mean Absolute Percentage Error (MAPE) of 5.32% and coefficient of determination (r²) of 0.9869, significantly outperforming traditional BP (25.22% MAPE) and GA-BP (13.61% MAPE) models. The model maintains robust accuracy during peak seasons (6.00% MAPE) and adverse weather conditions (5.50% MAPE), while also surpassing LSTM (8.20% MAPE) and Random Forest (9.80% MAPE) approaches. This advancement provides tourism managers with more reliable tools for visitor flow prediction, particularly in ecological sensitive areas like Banliang Ancient Village, contributing to sustainable tourism development and effective resource management.