Prediksi Penjualan Aerosol Menggunakan Algoritma ARIMA, LSTM Dan GRU
Abstract
Translated to English:
The advancement of information technology has significantly enhanced operational efficiency in companies through information systems that support data processing and more accurate decision-making. In the context of global competition, distributors face major challenges, such as shorter product life cycles and fluctuating customer demand. These dynamics directly impact inventory and production management, thereby increasing the need for more accurate forecasting solutions to support optimal production planning.
This study aims to compare the performance of ARIMA, LSTM, and GRU models in forecasting sales using time series data. ARIMA represents a traditional statistical approach, while LSTM and GRU are deep learning-based models capable of capturing more complex data patterns. The models were evaluated using four performance metrics: Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE).
The results indicate that LSTM outperformed the other models, achieving the lowest MAPE at 10.76%, followed by ARIMA at 11.23%, and GRU at 11.47%. LSTM demonstrated superior capability in capturing long-term trends and seasonal patterns, whereas GRU delivered nearly comparable performance with the added benefit of shorter training time. Although ARIMA is simpler and easier to implement, it was less effective in handling non-linear patterns in the data.
These findings offer valuable insights for companies in selecting the most appropriate forecasting model to optimize their supply chain, enhance operational efficiency, and support more informed decision-making.
Original Abstract:
Perkembangan teknologi informasi telah mendukung efisiensi operasional perusahaan melalui sistem informasi yang membantu dalam pengolahan data dan pengambilan keputusan yang lebih akurat. Dalam persaingan global, distributor menghadapi tantangan besar, seperti siklus hidup produk yang semakin pendek dan fluktuasi permintaan pelanggan. Hal ini berdampak pada pengelolaan stok dan produksi, yang menuntut solusi prediksi yang lebih akurat untukmendukung perencanaan produksi yang optimal. Penelitian ini bertujuan untuk membandingkan performa ARIMA, LSTM, dan GRU dalam memprediksi penjualan menggunakan metode time series forecasting. ARIMA merupakan pendekatan statistik tradisional, sedangkan LSTM dan GRU berbasis deep learning, yang mampu menangkap pola data yang lebih kompleks. Modeldievaluasi menggunakan metrik MSE, RMSE, MAE, dan MAPE. Hasil menunjukkan bahwa LSTM memberikan performa terbaik dengan MAPE 10,76%, diikuti oleh ARIMA (11,23%) dan GRU (11,47%). LSTM terbukti unggul dalam menangkap tren jangka panjang dan pola musiman, sementara GRU menunjukkan kinerja yang hampir setara dengan LSTM namun dengan waktu pelatihan yang lebih singkat. ARIMA, meskipun lebih sederhana, kurang efektif dalam menangani pola non-linear dalam data. Temuan ini memberikan wawasan bagi perusahaan dalammemilih model prediksi yang paling sesuai untuk mengoptimalkan rantai pasokan, meningkatkan efisiensi operasional, serta mendukung pengambilan keputusan yang lebih baik.
@ARTICLE{Sunendar2025-nf,
title = "Prediksi Penjualan Aerosol Menggunakan Algoritma ARIMA, LSTM
Dan GRU",
author = "Sunendar, Nendi and Putro, Harjono P and Hesananda, Rizki",
abstract = "The advancement of information technology has significantly
enhanced operational efficiency by enabling companies to process
data more effectively and make better decisions. In a highly
competitive global market, distributors face major challenges,
including shorter product life cycles and fluctuating customer
demand. These factors impact stock and production management,
necessitating more accurate predictive solutions to optimize
production planning. This study aims to compare the performance
of ARIMA, LSTM, and GRU models in sales forecasting using time
series forecasting methods. ARIMA represents a traditional
statistical approach, while LSTM and GRU, based on deep
learning, are capable of capturing complex data patterns. The
models are evaluated using MSE, RMSE, MAE, and MAPE metrics. The
results indicate that LSTM outperforms other models with a MAPE
of 10.76\%, followed by ARIMA (11.23\%) and GRU (11.47\%). LSTM
excels in identifying long-term trends and seasonal patterns,
while GRU achieves nearly comparable accuracy with a shorter
training time. ARIMA, despite its simplicity, struggles to
handle non-linear patterns. These findings provide valuable
insights for companies in selecting the most suitable predictive
model to optimize supply chain management, enhance operational
efficiency, and support more informed decision-making.",
journal = "INSOLOGI: Jurnal Sains dan Teknologi",
publisher = "Yayasan Literasi Sains Indonesia",
volume = 4,
number = 1,
pages = "113--126",
month = feb,
year = 2025
}