BACKPROPAGATION AND GARSON ALGORITHMS FOR IDENTIFYING DETERMINANTS AND PREDICTING FOREIGN TOURIST ARRIVALS IN MEDAN

Nuramal Linda; Ismail Husein; R. Maisaroh Rezyekiyah Siregar; Klause Roder; Abul Hashem Beg

doi:10.53806/jmscowa.v7i1.1254

Authors

Nuramal Linda Department of Mathematics, Universitas Islam Negeri Sumatera Utara, Medan, Indonesia
Ismail Husein Department of Mathematics, Universitas Islam Negeri Sumatera Utara, Medan, Indonesia
R. Maisaroh Rezyekiyah Siregar Department of Mathematics, Universitas Islam Negeri Sumatera Utara, Medan, Indonesia
Klause Roder Department of Applied Mathematics, Universty of Ausburg, Ausburg, Germany
Abul Hashem Beg La Trobe University, Australia

DOI:

https://doi.org/10.53806/jmscowa.v7i1.1254

Keywords:

Artificial Neural Networks (ANN); Backpropagation Algorithm; Garson Algorithms; Tourist Forecasting.

Abstract

Global anomalies cause nonlinear tourism volatility, which typical linear models struggle to accommodate. To overcome these difficulties, this study uses a deep [32-16-8-4] backpropagation neural network (BPNN) to anticipate international visitor arrivals in Medan from 2015 to 2024. Methodologically, the model incorporates pandemic-era data to successfully capture severe volatility, and Garson's technique is used to partially address the ANN 'black-box' limitation. The optimized BPNN fared better than typical linear regression benchmarks (>26% MAPE), with a mean absolute percentage error (MAPE) of 18% and a R2 of 0.3512. The weight split indicated hotel occupancy (42%) and currency rates (27%) as the key demand drivers, with a continuous recovery trend expected through December 2025. These data-driven findings offer regional authorities empirical justification to strategically augment peak-season hotel capacity by 15–20% and collaborate on sustaining currency stability, thereby transforming intricate neural learning into actionable, sustainable tourism development.