Prediction of Domestic Tourist Trips in North Sumatra Using Multilayer Perceptron Neural Networks (MLP)

Abstract

The tourism sector significantly contributes to the economic development of North Sumatra Province. To assist effective planning and policy decisions, reliable predictions of domestic tourist trips are needed. This research applies an Artificial Neural Network (ANN) using a Multilayer Perceptron (MLP) structure to estimate the number of domestic tourist visits to North Sumatra during the period 2019–2024. Considering that tourism data tend to be nonlinear and highly dynamic, the MLP approach was selected because of its ability to learn complex time-based patterns. A sliding window approach was applied, using 12 months as input variables and one month as the target for each forecast. Data preprocessing was conducted using Min–Max normalization to enhance learning performance. The network was trained with the Adam optimization algorithm for 2000 iterations. Several MLP model configurations with different numbers of hidden layers and neurons were compared to identify the most suitable structure. Model evaluation was carried out using Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE), and forecasting was tested over a 24-month period. Among the tested architectures, MLP (12–10–6) achieved the best performance with RMSE = 1,546,774.05, followed by MLP (12–6–2) with RMSE = 1,587,366.76, and MLP (12–32–16) with RMSE = 1,699,310.07. This study demonstrates that ANN, specifically MLP, is a robust tool for predicting tourism demand, offering actionable insights for stakeholders in North Sumatra’s tourism sector to guide sustainable development and resource allocation.