Comparative Analysis of Deep Learning Models for Forecasting Visitors at Mount Rinjani

This study compares five machine learning models: RNN, CNN, FFNN, LSTM, and ANFIS for forecasting daily visitors at Mount Rinjani National Park, addressing the need for accurate tourism demand prediction in protected areas. Using 1,650 observations from January 2021 to July 2025, model performance are evaluated with eight input features including temporal and autoregressive variables. Hyperparameter optimization was performed using grid search with a 75:25 train-test split. Results demonstrate LSTM's superior performance (MAPE 9.45%, RMSE 72.51, MAE 42.42), significantly outperforming RNN (14.30%), CNN (17.82%), FFNN (11.89%), and ANFIS (32.04%). The optimal LSTM configuration utilized a 30-day lookback window, 64 hidden units, 0.2 dropout rate, and 0.001 learning rate. LSTM's effectiveness in capturing long-term dependencies through gating mechanisms makes it highly suitable for tourism forecasting, offering practical insights for visitor management and resource planning.