Forecasting Jumlah Pendakian Gunung Panderman-Buthak Menggunakan Metode Long Short-Term Memory (LSTM) dalam Optimalisasi Pengelolaan

Mountain climbing is one of the most popular activities in Indonesia. This activity
offers an experience of exploration while enjoying the beauty of nature. However,
managers will face risks if climbers do not have adequate knowledge until
environmental pollution can reduce tourism appeal. The number of climbs also
varies with the seasons, increasing in the dry season and decreasing in the rainy
season. Natural disasters can also affect income. This study proposes the use of the
Long Short Term Memory (LSTM) algorithm to predict the number of climbs based
on entrance ticket sales data, because the processed data is time series data. The
obtained data consists of the daily recapitulation results of Mount Panderman-
Buthak climbing ticket sales, which spans from the 1st of January 2022
until the 31st of December 2023. This dataset was obtained from Perum Perhutani
KPH Malang Asper Pujon with permission from Perum Perhutani East Java
Regional Division. The results of this study obtained the best evaluation values in
the 80:20 data composition scenario. This data composition scenario has an RMSE
value of 0.749 in the train data and 1.446 in the test data. Meanwhile, the MAPE
value in the train data is 1.2% and 0.7% in the test data. This evaluation result
utilizes the grid search technique in determining the optimal hyperparameters. The
LSTM model is able to learn the train data pattern stably without significant
differences between train loss and test loss in the accuracy graph. This data
composition scenario uses 500 epochs, a batch size of 32, 256 neurons, and a
dropout rate of 20% with a validation score from the grid search technique of
99.9%.