Predicting the Number of Thai Tourists Visiting Indonesia Per Month Using Multi-Layer Perceptron Backpropagation and Adam

Abstract

This research seeks to forecast the monthly number of Thai tourist arrivals in Indonesia using an Artificial Neural Network (ANN) model based on Multilayer Perceptron (MLP) and the backpropagation algorithm. The historical tourist data is treated as a time series, applying the sliding window technique to use data from the past 12 months as input features and the following month as the target output. To enhance the model's training stability, the data is normalized using Min–Max Scaling. The constructed MLP model includes two hidden layers, with the input layer consisting of 12 neurons, each hidden layer containing 21 neurons, and a single neuron in the output layer. The first hidden layer utilizes the ReLU (Rectified Linear Unit) activation function, while the second layer employs tanh (hyperbolic tangent) to capture more intricate nonlinear patterns. The model is trained using the Adam optimizer with a learning rate of 0.01 over 2000 iterations, aiming to minimize the Mean Squared Error (MSE). The training results indicate that the MLP model effectively learns the data patterns, providing accurate predictions with strong alignment between the predicted and actual values. Predictions for the next 24 months reveal a fluctuating trend in tourist visits, mirroring the dynamic characteristics of time series data. Therefore, this approach proves to be an effective tool for tourism policy formulation and decision-making within the Indonesian tourism industry. The Adam optimizer and tanh activation function contribute to the model's enhanced learning efficiency and stability, especially when handling data with significant fluctuations.

Keywords: Forecasting, Tourists, Thailand, Indonesia, Arrivals, Tourism, Data, Artificial Neural Network, Multilayer Perceptron, Backpropagation, Adam, ReLU, tanh.