<![CDATA[This study compares the performance of two deep learning architectures—Multilayer Perceptron (MLP) and Long Short-Term Memory (LSTM)—for daily demand forecasting on an e-commerce delivery platform. The dataset consists of 1,827 daily observations from 2020 to 2024 and includes operational, temporal, and behavioral features such as holiday indicators, promotion signals, active customers, and delivery time. Data preprocessing includes cleaning, feature engineering, scaling, and sequence generation using a 30-day sliding window. Both models were trained and evaluated using consistent experimental settings and performance metrics. The results show that the LSTM model achieves better accuracy than the MLP model, with an RMSE of 811.81 compared to 830.15, while the difference in MAE between the two models remains minimal. LSTM demonstrates superior capability in capturing temporal dependencies and reacting to rapid demand fluctuations, whereas both models face challenges when predicting sudden demand spikes. These findings indicate that memory-based models such as LSTM are more effective for highly volatile time-series forecasting in e-commerce operations. However, performance can be further improved with the addition of external variables such as real-time promotions, weather conditions, and multivariate features.]]>
Copyrights © 2026
