Zirai Kimyasal Ürünler İçin Talep Tahmininin İleri Analitik Yöntemlerle Geliştirilmesi

Demand forecasting is an essential part of supply chain planning. Accurate prediction of sales directly affect the resource efficiency and success of inventory management process. However it is especially challenging for some business domains, including agriculture, due to strong seasonal patterns, fluctuating demand drivers, and market uncertainty. This study searches for the enhancement of demand forecasting for agrochemical products by applying machine learning and deep learning algorithms, including multi-series LSTM; 1-D multi-series CNN; and Prophet. The results are compared against traditional statistical model SARIMA as a baseline. Effects of diverse economic and environmental indicators on prediction accuracy are evaluated by experiments. Additionally, this study searches for the impact of the aggregation and disaggregation method for the prediction accuracy of intermittent time series. Model performance was evaluated across multiple error metrics for different products. The findings highlight that modeling algorithms can compete with human predictions and can improve demand forecasting process.