<![CDATA[The data sharing stage is an important step in model building using Artificial Neural Network (ANN) methods to avoid the risk of overfitting and underfitting that can affect model performance. Proper data division aims to ensure that the model can generalize well to data that has never been seen before. Generally, data sharing is done by dividing the dataset into two main parts, namely training and testing data. However, to better address overfitting, there are also those who divide the data into three parts, namely training, testing, and validation. This study aims to evaluate the performance of ANN modelling using these two ways of dividing data. The model is evaluated using Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE) metrics to measure prediction error. The data used is data on foreign tourist arrivals to Indonesia, which has a fluctuating pattern and is influenced by calendar effects. The results show that the data division type with two groups generally produces a smaller MAPE value than the data division into three groups. However, the model with two parts of data is not able to capture the seasonal pattern in the data. On the other hand, the model with three parts of data can overcome this problem better. The best model was obtained with the proportion of training data, validation data, and test data of 80%, 10%, and 10%, respectively, which resulted in a MAPE value of 24.45%.]]>
Copyrights © 2025
