Publications-Theses
Article View/Open
Publication Export
-
Google ScholarTM
NCCU Library
Citation Infomation
Related Publications in TAIR
題名 從實務角度發展機器學習演算法的需求預測模型
Developing a Demand Forecasting Model by Machine Learning Algorithms: A Practical Approach作者 歐陽鈵鈞
Ou Yang, Bing-Jyun貢獻者 洪叔民
Horng, Shwu-Min
歐陽鈵鈞
Ou Yang, Bing-Jyun關鍵詞 需求預測
機器學習
統計方法
供應鏈管理
電商銷量預測
庫存規劃
Demand Forecasting
Machine Learning
Statistics
Supply Chain Management
E-commerce sales forecasting
Inventory Planning日期 2024 上傳時間 4-Aug-2025 13:41:45 (UTC+8) 摘要 本研究旨在探討需求預測方法，通過協助一家全球消費性電子硬體品牌商建立機器學習模型，對比分析機器學習模型與統計方法在預測具有時間序列特徵的產品銷售量方面的表現，以期為企業優化需求預測流程提供參考。研究中，我們利用個案企業提供的歷史銷售數據，構建了包括統計方法和機器學習模型在內的預測方法。 統計方法涵蓋了企業現有的移動平均法和指數平滑法，並引入了多變量迴歸、ARIMA 及 SARIMA 模型。而在機器學習方面，我們則採用了 XGBoost 和 LSTM 演算法。所使用的數據集為2020年8月1日至2023年8月1日三年期的歷史銷售數據。研究中考慮的訓練變數包括某類產品的「銷售量」、「售價」、「折扣」、「重要的節日和活動日」及「時間特徵值」。我們將2023年5月1日至2023年8月1日的銷量作為測試數據集，並以2023年5月1日之前的數據作為訓練和驗證數據集。在 XGBoost 和 LSTM 模型上，我們使用了五折交叉驗證，並計算平均均方誤差（MSE）來評估訓練結果。 研究結果表明，對於具有季節性銷售離群值的銷售數據，混合預測方法的效果優於單一模型。在處理離群值時，我們使用移動平均法預測其季節性變化，對其餘數據則使用 XGBoost 進行預測。測試數據集的結果顯示，與其他方法相比，混合堆疊模型達到了最小的 MSE，證明其在預測精度上具有明顯的優勢。此外，通過與個案企業在建模過程中的討論和訪談，我們分析了企業在引入機器學習模型以優化現有需求預測流程時需要考慮的因素。這些見解將為未來有相關需求的企業提供指導，進而提升建模和應用的成功率。
This study aims to explore demand forecasting methods by assisting a global consumer electronics hardware brand in building machine learning models and comparing the performance of machine learning models and statistical methods in forecasting product sales with time series characteristics. The objective is to provide a reference for companies to optimize their demand forecasting processes. In this research, we utilized historical sales data provided by the case company to construct both statistical forecasting methods and machine learning models. The statistical methods include the company’s existing moving average method and exponential smoothing, as well as multivariate regression, ARIMA, and SARIMA models. On the machine learning side, we adopted XGBoost and LSTM algorithms. The dataset used spans three years, from August 1, 2020, to August 1, 2023. The training variables considered in this study include "sales volume," "price," "discount," "important holidays and event days," and "time features" for a particular product category. We used the sales data from May 1, 2023, to August 1, 2023, as the test dataset and the data before May 1, 2023, as the training and validation dataset. In the XGBoost and LSTM models, we employed 5-fold cross-validation and calculated the average Mean Squared Error (MSE) to evaluate the training results. The findings of the study indicate that for sales data with seasonal outliers, a hybrid forecasting method performs better than a single model. In predicting outliers, we used the moving average method to forecast their seasonal variations, while the remaining data were predicted using XGBoost. The results on the test dataset showed that the hybrid stacking model achieved the lowest MSE compared to other methods, demonstrating its superior forecasting accuracy. Additionally, through discussions and interviews with the case company during the modeling process, we analyzed the factors that companies need to consider when introducing machine learning models to optimize their existing demand forecasting processes. These insights will provide guidance for other companies with similar needs in the future, thereby improving the success rate of modeling and implementation.參考文獻 英文文獻 Alpaydin, E. (2020). Introduction to Machine Learning. Massachusetts Institute of Technology Athanasiadis, C. L., Tsoumplekas, G., Chrysopoulos, A., & Doukas, D. I. (2022). Peak demand forecasting: A comparative analysis of state-of-the-art machine learning techniques. IEEE, 22. Chatzis, S. P., Siakoulis, V., Petropoulos , A., Stavroulakis, E., & Vlachogiannakis, Nikos. (2018). Machine learning that matters. Forecasting stock market crisis events using deep and statistical machine learning techniques, 112, 353–371. https://doi.org/https://doi.org/10.1016/j.eswa.2018.06.032 Chen, T. & Guestrin, C. (2016). XGBoost: A scalable tree boosting system. KDD, 16,785-794. Barochiner, D., Lado, R., Carletti, L., & Pinter, F. (2022). A machine learning approach to address 1-week-ahead peak demand forecasting using the XGBoost algorithm. IEEE, 22. Delua, J. (2021). Supervised versus unsupervised learning: What's the difference. IBM. https://www.ibm.com/blog/supervised-vs-unsupervised-learning/ Hodzic, K., Hasic, H., Cogo, E. & Juric, Z. (2019). Warehouse demand forecasting based on long short-term memory neural networks. IEEE, 19. Hou, L., Shen, X., & Zhang, L. (2023). Urban electricity demand forecasting with a hybrid machine learning model. ICNSC, 23. Jordan, I. M., & Mitchell, M. T. (2015). Machine learning: Trends, perspectives, and prospects. SCIENCE, 349(6245), 255–260. Krishna, K. S. R., Pasula, P., Kavyakeerthi, T., & Karthik, I. (2022). Identifying demand forecasting using machine learning for business intelligence. ICCMC, 938-942 Nilsson, N. J. (1982). Principles of Artificial Intelligence. Springer-Verlag. Ma Z., Zhang, Z. & Wang, C. (2021). Deep learning algorithms for automotive spare parts demand forecasting. CISAI, 21, 358-361. Mediavilla, M. A., Dietrich, F., & Palm, D. (2022). Review and analysis of artificial intelligence methods for demand forecasting in supply chain management. CIRP, 1126-1131. Menon, S., Sultanova, N., Jayabalan, M., & Mustafina, J. (2023). A study on data-driven energy forecasting: a machine learning perspective. DeSE, 23, 702-705. Shabbir, N., Kutt, L., Raja, H. A., Ahmadiahangar, R., Rosin, A., & Husev, O. (2021). Machine learning and deep learning techniques for residual load forecasting: a comparative analysis. RTUCON. Sharma, A. K., Kiran, M., Pauline, S. J. P., Maheshwari, P. & Divakar V. (2021). Demand forecasting using coupling of machine learning and time series models for the automotive aftermarket sector. ICEECCOT. 10-11, 832-836. Shi, Y., Zhang, W., Sang, F. M., Zhao, L. & Wang, T. (2022). Application of ALO-ELM on electricity demand forecasting under the spot power market. SPIES, 22, 2194-2197. Steel, R. G. D.& Torrie, J. H. (1960). Principles and Procedures of Statistics. McGraw-Hill Book Company. Thakur, A., Shukla, K. A., Choudhary, A. & Atrey, J. (2023). Predictive analysis of energy consumption and electricity demand using machine learning techniques. ICSSES, 23. Turing, A. M. (1950). Computing machinery and intelligence. Mind, 59, 433-460. Wagstaff, K. L. (2012). Machine learning that matters. Proceedings of the Twenty-Ninth International Conference on Machine Learning (ICML), 529–536. https://doi.org/https://doi.org/10.48550/arXiv.1206.4656 中文文獻 何宗武 (2014)。《追蹤資料分析：原理與R程式實務》。台灣：雙葉書廊。 馮晨與陳志德 (2019)。〈追蹤資料分析：原理與R程式實務〉。《計算機系統應用》，28，226-232。 描述 碩士
國立政治大學
企業管理研究所(MBA學位學程)
111363065資料來源 http://thesis.lib.nccu.edu.tw/record/#G0111363065 資料類型 thesis dc.contributor.advisor 洪叔民 zh_TW dc.contributor.advisor Horng, Shwu-Min en_US dc.contributor.author (Authors) 歐陽鈵鈞 zh_TW dc.contributor.author (Authors) Ou Yang, Bing-Jyun en_US dc.creator (作者) 歐陽鈵鈞 zh_TW dc.creator (作者) Ou Yang, Bing-Jyun en_US dc.date (日期) 2024 en_US dc.date.accessioned 4-Aug-2025 13:41:45 (UTC+8) - dc.date.available 4-Aug-2025 13:41:45 (UTC+8) - dc.date.issued (上傳時間) 4-Aug-2025 13:41:45 (UTC+8) - dc.identifier (Other Identifiers) G0111363065 en_US dc.identifier.uri (URI) https://nccur.lib.nccu.edu.tw/handle/140.119/158417 - dc.description (描述) 碩士 zh_TW dc.description (描述) 國立政治大學 zh_TW dc.description (描述) 企業管理研究所(MBA學位學程) zh_TW dc.description (描述) 111363065 zh_TW dc.description.abstract (摘要) 本研究旨在探討需求預測方法，通過協助一家全球消費性電子硬體品牌商建立機器學習模型，對比分析機器學習模型與統計方法在預測具有時間序列特徵的產品銷售量方面的表現，以期為企業優化需求預測流程提供參考。研究中，我們利用個案企業提供的歷史銷售數據，構建了包括統計方法和機器學習模型在內的預測方法。 統計方法涵蓋了企業現有的移動平均法和指數平滑法，並引入了多變量迴歸、ARIMA 及 SARIMA 模型。而在機器學習方面，我們則採用了 XGBoost 和 LSTM 演算法。所使用的數據集為2020年8月1日至2023年8月1日三年期的歷史銷售數據。研究中考慮的訓練變數包括某類產品的「銷售量」、「售價」、「折扣」、「重要的節日和活動日」及「時間特徵值」。我們將2023年5月1日至2023年8月1日的銷量作為測試數據集，並以2023年5月1日之前的數據作為訓練和驗證數據集。在 XGBoost 和 LSTM 模型上，我們使用了五折交叉驗證，並計算平均均方誤差（MSE）來評估訓練結果。 研究結果表明，對於具有季節性銷售離群值的銷售數據，混合預測方法的效果優於單一模型。在處理離群值時，我們使用移動平均法預測其季節性變化，對其餘數據則使用 XGBoost 進行預測。測試數據集的結果顯示，與其他方法相比，混合堆疊模型達到了最小的 MSE，證明其在預測精度上具有明顯的優勢。此外，通過與個案企業在建模過程中的討論和訪談，我們分析了企業在引入機器學習模型以優化現有需求預測流程時需要考慮的因素。這些見解將為未來有相關需求的企業提供指導，進而提升建模和應用的成功率。 zh_TW dc.description.abstract (摘要) This study aims to explore demand forecasting methods by assisting a global consumer electronics hardware brand in building machine learning models and comparing the performance of machine learning models and statistical methods in forecasting product sales with time series characteristics. The objective is to provide a reference for companies to optimize their demand forecasting processes. In this research, we utilized historical sales data provided by the case company to construct both statistical forecasting methods and machine learning models. The statistical methods include the company’s existing moving average method and exponential smoothing, as well as multivariate regression, ARIMA, and SARIMA models. On the machine learning side, we adopted XGBoost and LSTM algorithms. The dataset used spans three years, from August 1, 2020, to August 1, 2023. The training variables considered in this study include "sales volume," "price," "discount," "important holidays and event days," and "time features" for a particular product category. We used the sales data from May 1, 2023, to August 1, 2023, as the test dataset and the data before May 1, 2023, as the training and validation dataset. In the XGBoost and LSTM models, we employed 5-fold cross-validation and calculated the average Mean Squared Error (MSE) to evaluate the training results. The findings of the study indicate that for sales data with seasonal outliers, a hybrid forecasting method performs better than a single model. In predicting outliers, we used the moving average method to forecast their seasonal variations, while the remaining data were predicted using XGBoost. The results on the test dataset showed that the hybrid stacking model achieved the lowest MSE compared to other methods, demonstrating its superior forecasting accuracy. Additionally, through discussions and interviews with the case company during the modeling process, we analyzed the factors that companies need to consider when introducing machine learning models to optimize their existing demand forecasting processes. These insights will provide guidance for other companies with similar needs in the future, thereby improving the success rate of modeling and implementation. en_US dc.description.tableofcontents 謝誌 I 中文摘要 II 英文摘要 III 目次 IV 表次 V 圖次 VI 第一章 緒論 1 1.1 研究背景與動機 1 1.2 研究目的 2 1.3 研究流程 3 第二章 文獻探討 6 2.1 統計預測方法 6 2.2 機器學習預測方法 13 第三章 研究方法 19 3.1 研究架構 19 3.2 研究方法 22 3.3 研究設計 23 第四章 研究結果 25 4.1 模型成效評估 25 4.2 模型建構的流程 31 第五章 結論與建議 34 5.1 結論 34 5.2 建議 35 5.3 研究限制 35 參考文獻 37 zh_TW dc.format.extent 1899763 bytes - dc.format.mimetype application/pdf - dc.source.uri (資料來源) http://thesis.lib.nccu.edu.tw/record/#G0111363065 en_US dc.subject (關鍵詞) 需求預測 zh_TW dc.subject (關鍵詞) 機器學習 zh_TW dc.subject (關鍵詞) 統計方法 zh_TW dc.subject (關鍵詞) 供應鏈管理 zh_TW dc.subject (關鍵詞) 電商銷量預測 zh_TW dc.subject (關鍵詞) 庫存規劃 zh_TW dc.subject (關鍵詞) Demand Forecasting en_US dc.subject (關鍵詞) Machine Learning en_US dc.subject (關鍵詞) Statistics en_US dc.subject (關鍵詞) Supply Chain Management en_US dc.subject (關鍵詞) E-commerce sales forecasting en_US dc.subject (關鍵詞) Inventory Planning en_US dc.title (題名) 從實務角度發展機器學習演算法的需求預測模型 zh_TW dc.title (題名) Developing a Demand Forecasting Model by Machine Learning Algorithms: A Practical Approach en_US dc.type (資料類型) thesis en_US dc.relation.reference (參考文獻) 英文文獻 Alpaydin, E. (2020). Introduction to Machine Learning. Massachusetts Institute of Technology Athanasiadis, C. L., Tsoumplekas, G., Chrysopoulos, A., & Doukas, D. I. (2022). Peak demand forecasting: A comparative analysis of state-of-the-art machine learning techniques. IEEE, 22. Chatzis, S. P., Siakoulis, V., Petropoulos , A., Stavroulakis, E., & Vlachogiannakis, Nikos. (2018). Machine learning that matters. Forecasting stock market crisis events using deep and statistical machine learning techniques, 112, 353–371. https://doi.org/https://doi.org/10.1016/j.eswa.2018.06.032 Chen, T. & Guestrin, C. (2016). XGBoost: A scalable tree boosting system. KDD, 16,785-794. Barochiner, D., Lado, R., Carletti, L., & Pinter, F. (2022). A machine learning approach to address 1-week-ahead peak demand forecasting using the XGBoost algorithm. IEEE, 22. Delua, J. (2021). Supervised versus unsupervised learning: What's the difference. IBM. https://www.ibm.com/blog/supervised-vs-unsupervised-learning/ Hodzic, K., Hasic, H., Cogo, E. & Juric, Z. (2019). Warehouse demand forecasting based on long short-term memory neural networks. IEEE, 19. Hou, L., Shen, X., & Zhang, L. (2023). Urban electricity demand forecasting with a hybrid machine learning model. ICNSC, 23. Jordan, I. M., & Mitchell, M. T. (2015). Machine learning: Trends, perspectives, and prospects. SCIENCE, 349(6245), 255–260. Krishna, K. S. R., Pasula, P., Kavyakeerthi, T., & Karthik, I. (2022). Identifying demand forecasting using machine learning for business intelligence. ICCMC, 938-942 Nilsson, N. J. (1982). Principles of Artificial Intelligence. Springer-Verlag. Ma Z., Zhang, Z. & Wang, C. (2021). Deep learning algorithms for automotive spare parts demand forecasting. CISAI, 21, 358-361. Mediavilla, M. A., Dietrich, F., & Palm, D. (2022). Review and analysis of artificial intelligence methods for demand forecasting in supply chain management. CIRP, 1126-1131. Menon, S., Sultanova, N., Jayabalan, M., & Mustafina, J. (2023). A study on data-driven energy forecasting: a machine learning perspective. DeSE, 23, 702-705. Shabbir, N., Kutt, L., Raja, H. A., Ahmadiahangar, R., Rosin, A., & Husev, O. (2021). Machine learning and deep learning techniques for residual load forecasting: a comparative analysis. RTUCON. Sharma, A. K., Kiran, M., Pauline, S. J. P., Maheshwari, P. & Divakar V. (2021). Demand forecasting using coupling of machine learning and time series models for the automotive aftermarket sector. ICEECCOT. 10-11, 832-836. Shi, Y., Zhang, W., Sang, F. M., Zhao, L. & Wang, T. (2022). Application of ALO-ELM on electricity demand forecasting under the spot power market. SPIES, 22, 2194-2197. Steel, R. G. D.& Torrie, J. H. (1960). Principles and Procedures of Statistics. McGraw-Hill Book Company. Thakur, A., Shukla, K. A., Choudhary, A. & Atrey, J. (2023). Predictive analysis of energy consumption and electricity demand using machine learning techniques. ICSSES, 23. Turing, A. M. (1950). Computing machinery and intelligence. Mind, 59, 433-460. Wagstaff, K. L. (2012). Machine learning that matters. Proceedings of the Twenty-Ninth International Conference on Machine Learning (ICML), 529–536. https://doi.org/https://doi.org/10.48550/arXiv.1206.4656 中文文獻 何宗武 (2014)。《追蹤資料分析：原理與R程式實務》。台灣：雙葉書廊。 馮晨與陳志德 (2019)。〈追蹤資料分析：原理與R程式實務〉。《計算機系統應用》，28，226-232。 zh_TW
