The current competitive manufacturing environment drives many companies to respond quickly to demand. One effective approach is to predict future market conditions. In this research, sales predictions were carried out using a case study of a lunch box manufacturing company. The company requires a method to predict lunch box sales to estimate the number of products to be produced. This aims to prevent excessive overproduction or underproduction. The prediction methods used in this research were Long Short-Term Memory (LSTM) and Autoregressive Integrated Moving Average (ARIMA). The LSTM method tends to be better suited to non-linear data such as market conditions, while ARIMA was used for comparison. Based on the prediction results for the company\'s two products, the LSTM method performed better than ARIMA in all assessment types: Mean Absolute Deviation (MAD), Mean Absolute Percentage Error (MAPE), and Mean Squared Error (MSE).
Introduction
In a fast-paced manufacturing environment, accurate sales forecasting is critical for inventory management, production planning, and financial decision-making.
Traditional statistical models like ARIMA and modern deep learning models like LSTM are widely used for time-series forecasting.
While ARIMA excels in capturing linear patterns, it struggles with the non-linear and chaotic nature of real-world sales data.
LSTM (Long Short-Term Memory) networks are better suited to modeling complex, non-linear, and long-term dependencies in sequential data.
2. Literature Review
Early forecasting models: Exponential smoothing, Moving Averages, ARIMA.
ARIMA is based on the Box-Jenkins methodology with steps: model identification (p, d, q), parameter estimation, model validation, and forecasting.
However, ARIMA struggles with non-linear patterns.
LSTM, a type of RNN, overcomes the vanishing gradient problem and captures long-term dependencies using memory cells and gates (input, forget, output).
LSTM is especially effective for complex sequences like sales data.
3. Methodology
ARIMA Forecasting Steps:
Test for stationarity (ADF test).
Tune parameters (p, d, q) using combinations or ACF/PACF plots.
Split data into train, validation, and test.
Use the best parameters for forecasting and evaluate using:
MAD (Mean Absolute Deviation)
MAPE (Mean Absolute Percentage Error)
MSE (Mean Squared Error)
LSTM Forecasting Steps:
Normalize data (MinMaxScaler).
Use a sliding window to create sequences.
LSTM model architecture:
Dense layers (64 and 32 neurons), dropout layer.
Optimizer: Adam, Loss: MSE, Epochs: 100.
Use multi-step prediction and evaluate with the same error metrics.
4. Case Study: Lunch Box Manufacturing Company
Dataset: Monthly sales data (2020–2022) for two products:
7200ml sealware
3500ml sealware
ARIMA Results:
Parameters:
7200ml: (6,1,5)
3500ml: (5,1,7)
Average metrics:
MAD: 131.72
MAPE: 38.08%
MSE: 26,715.37
LSTM Results:
Trained on the same dataset with a multi-step forecasting approach.
Average metrics:
MAD: 76.55
MAPE: 21.26%
MSE: 10,470.13
Conclusion
This research predicted the sales of lunch boxes produced by a company using ARIMA and LSTM methods. The experiment was conducted using a dataset of sales data for two years. From the experimental results and measurements using MAD, MAPE, and MSE, it was shown that the prediction results using LSTM had better performance than the results using ARIMA. This shows that the LSTM method is more suitable for use on datasets for this case study because the lunch box sales data is non-linear.
References
[1] Menculini, L., Marini, A., Proietti, M., Garinei, A., Bozza, A., Moretti, C., & Marconi, M., “Comparing Prophet and Deep Learning to ARIMA in Forecasting Wholesale Food Prices,” Forecasting, vol. 3, no. 3, pp. 644-662, 2021. https://doi.org/10.3390/forecast3030040.
[2] Ghienne, M., & Limare, A., “Learning structural stress virtual sensors from on-board instrumentation of a commercial aircraft,” Computers & Structures, vol. 289, 107155, 2023. https://doi.org/10.1016/j.compstruc.2023.107155.
[3] Zhou, D., Zhuang, X., Zuo, H., Wang, H., & Yan, H., “Deep Learning-Based Approach for Civil Aircraft Hazard Identification and Prediction,” IEEE Access, vol. 8, pp. 103665-103683, 2020. https://doi.org/10.1109/ACCESS.2020.2997371.
[4] Siami?Namini, S., & Namin, A. S., “Forecasting Economics and Financial Time Series: ARIMA vs. LSTM,” arXiv (Cornell University), 2018. https://doi.org/10.48550/arxiv.1803.06386.
[5] Alamu, O. S., & Siam, M. K., “Stock Price Prediction and Traditional Models: An Approach to Achieve Short, Medium and Long-Term Goals,” Journal of Intelligent Learning Systems and Applications, vol. 16, no. 4, pp. 363-383, 2024. https://doi.org/10.4236/jilsa.2024.164018.
[6] Verano, D. A., Husnawati, H., & Ermatita, E., “Implementation of Autoregressive Integrated Moving Average Model to Forecast Raw Material Stock in The Digital Printing Industry,” Journal of Information Technology and Computer Science, vol. 5, no. 1, 2020. https://doi.org/10.25126/jitecs.202051117.
[7] Kontopoulou, V. I., Panagopoulos, A. D., Kakkos, I., & Matsopoulos, G. K., “A Review of ARIMA vs. Machine Learning Approaches for Time Series Forecasting in Data Driven Networks,” Future Internet, vol. 15, no. 8, 255, 2023. https://doi.org/10.3390/fi15080255.
[8] Hadwan, M., Al-Maqaleh, B. M., Al-Badani, F. N., Khan, R. U., & Al-Hagery, M. A., “A Hybrid Neural Network and Box-Jenkins Models for Time Series Forecasting,” Computers, Materials & Continua/Computers, Materials & Continua, vol. 70, no. 3, pp. 4829-4845, 2021.
https://doi.org/10.32604/cmc.2022.017824.
[9] Gasparin, A., Lukovi?, S., & Alippi, C., “Deep Learning for Time Series Forecasting: The Electric Load Case,” arXiv (Cornell University), 2019. https://doi.org/10.48550/arxiv.1907.09207.
[10] Islam, M. R., Zamil, M. Z. H., Rayed, M. E., Kabir, M. M., Mridha, M. F., Nishimura, S., & Shin, J., “Deep Learning and Computer Vision Techniques for Enhanced Quality Control in Manufacturing Processes,” IEEE Access vol. 12, pp. 121449-121479, 2024.
https://ieeexplore.ieee.org/abstract/document/10663422.
[11] Lee, S. I., & Yoo, S. J., “A Deep Efficient Frontier Method for Optimal Investments,” arXiv (Cornell University), 2017. https://arxiv.org/pdf/1709.09822.
[12] Albeladi, K., Zafar, B., & Mueen, “A. Time Series Forecasting using LSTM and ARIMA,” International Journal of Advanced Computer Science and Applications, vol. 14, no. 1, 2023. https://doi.org/10.14569/ijacsa.2023.0140133.
[13] Singh, D. A. A. G., Leavline, E. J., Muthukrishnan, S., & Yuvaraj, R., “Machine Learning based Business Forecasting,” International Journal of Information Engineering and Electronic Business, vol. 10, no. 6, pp. 40-51, 2018. https://doi.org/10.5815/ijieeb.2018.06.05.
[14] Skenderi, G., Joppi, C., Denitto, M., & Cristani, M., “On the Use of Learning-Based Forecasting Methods for Ameliorating Fashion Business Processes: A Position Paper,” In Lecture Notes in Computer Science vol. 13644, pp. 647-659, 2023. https://doi.org/10.1007/978-3-031-37742-6_50.
[15] Ganguly, P., & Mukherjee, I., “Enhancing Retail Sales Forecasting with Optimized Machine Learning Models,” 4th International Conference on Sustainable Expert Systems (ICSES), 2024. https://doi.org/10.1109/icses63445.2024.10762950.
[16] Ghienne, M., & Limare, A. “Learning structural stress virtual sensors from on-board instrumentation of a commercial aircraft,” Computers & Structures, vol. 289, 2023. https://doi.org/10.1016/j.compstruc.2023.107155
[17] Wang, J., Guo, Y., & Wang, Y., “A sequential random forest for short-term vessel speed prediction,” Ocean Engineering, vol. 248, 2022. https://doi.org/10.1016/j.oceaneng.2022.110691
[18] Abdulkadir, S. J., Alhussian, H., Nazmi, M. A. M., & Elsheikh, A. A. “Long Short Term Memory Recurrent Network for Standard and Poor’s 500 Index Modelling,” International Journal of Engineering & Technology, vol. 7, no, 25, 2018. https://doi.org/10.14419/ijet.v7i4.15.21365