LOGISTICS AND TRANSPORTATION: FREIGHT DEMAND FORECASTING USING SARIMA AND XGBOOST

LSTM-XGBoost hybrid neural network for multi-scenario inventory and multi-dimensional comprehensive demand forecasting in the supply chain

Due to the diverse range of product categories in e-commerce supply chains, precise prediction of warehouse inventory levels has become essential for improving resource allocation and minimizing operational expenses. This research delves into the modeling and analysis of inventory and sales performance across 350 product types within the supply chain framework. To enhance inventory forecasting accuracy, a multi-model evaluation system was developed, combining ARIMA, grey forecasting, and XGBoost regression techniques. Data preprocessing was rigorously validated using ADF tests for stationarity and residual white noise analysis. Based on a comparison of Mean Absolute Percentage Error (MAPE) metrics, the XGBoost model was chosen for its strong nonlinear modeling ability to forecast average monthly inventory levels over the next three months. For daily sales forecasting, the study utilized LSTM neural networks alongside the Prophet model for time series decomposition, allowing for the isolation of long-term trends, seasonal effects from holidays, and the impact of promotional events. A two-step calibration process enabled the model to effectively predict daily sales for the upcoming 90-day period. The findings suggest that the multi-model hyperparameter tuning strategy greatly improves the reliability of inventory forecasts, offering solid analytical backing for informed business decisions.

Comparative Analysis of Statistical and Machine Learning Models for Air Passenger Demand Forecasting: Evidence from Domestic and International Passenger Segments

Accurate forecasting of air passenger demand is critical for aviation sector planning, yet model selection remains contested when data exhibit structural discontinuities and segment-specific behavior. This study benchmarks four forecasting approaches—SARIMA, SARIMAX, Prophet, and XGBoost—on monthly domestic and international passenger traffic data from 2019 to 2024, using a 12-month hold-out test set and a 36-month projection horizon. Model performance was evaluated using MAE, RMSE, and MAPE, with stationarity confirmed via Augmented Dickey-Fuller tests.Results reveal stark segmentation in model suitability. For international passengers, SARIMA achieved the lowest test MAPE (8.66%), substantially outperforming XGBoost (14.23%) and Prophet (57.65%), attributable to the strong first-order autocorrelation structure of post-recovery international traffic. For domestic passengers, SARIMA and SARIMAX both exhibited explosive forecast divergence (MAPE exceeding 11,900%), rendering them operationally unusable; XGBoost (MAPE 14.87%) emerged as the most reliable alternative by capturing non-linear interactions among lag features, rolling statistics, and annual trend components.The central finding concerns the role of exogenous variables: SARIMAX produced metrics identical to SARIMA for international passengers and marginally worse for domestic passengers, yielding no incremental predictive gain in either segment. This challenges the assumption that exogenous augmentation inherently improves seasonal time series models—such benefits are conditional on the relevance, stability, and non-redundancy of external regressors relative to existing seasonal components. These findings offer empirical guidance for segment-specific model selection in aviation demand forecasting and caution against uncritical adoption of SARIMAX when core seasonal structure is already sufficient.

Fine-Grained Forecasting of Short-Haul Logistics Volume Based on Time Series Feature Engineering and XGBoost Algorithm

: Addressing the demand for high-frequency response efficiency in modern short-haul logistics networks, this study investigates fine-grained forecasting of short-haul logistics volume. First, historical logistics volume data undergoes preprocessing to screen target routes and eliminate outliers, while a custom function enables scientific segmentation of production periods across dates. Subsequently, periodic decomposition techniques were applied to decompose the raw time series into trend, seasonal, and residual components, confirming the pronounced cyclical patterns in logistics demand and the stationarity of residuals. During modeling, a sliding-window feature matrix was constructed to convert time-series data into supervised learning format. The XGBoost gradient boosting algorithm was employed, with the objective function optimized via second-order Taylor expansion to effectively capture nonlinear characteristics in freight volume sequences. The results show that the model achieves a coefficient of determination of 0.94 on the independent test set, with an average absolute percentage error of only 4%, demonstrating outstanding fitting capability and generalization performance. Ultimately, the model successfully achieves cargo volume extrapolation at 10-minute granularity for the next 24 hours, accurately capturing the bimodal characteristics of cargo volume on specific routes, laying a solid data foundation for subsequent dynamic scheduling optimization.

Application of Forecasting and Transportation Models to Optimize Beef Distribution Costs

Inaccurate demand estimation often leads to inefficient distribution planning and increased transportation costs, particularly in perishable food distribution systems. This study aims to optimize beef distribution costs by integrating forecasting techniques with transportation optimization methods. Historical sales data from three distribution destinations Baros, Cimahi, and Dewi Sartika were analyzed using several forecasting methods, namely Single Moving Average, Weighted Moving Average, Exponential Smoothing, and Linear Regression. Forecasting accuracy was evaluated using the Mean Absolute Deviation (MAD) criterion to determine the most suitable method for each destination. The results show that Linear Regression generated the lowest MAD value for Baros (4.77), while Exponential Smoothing with α = 0.1 produced the most accurate forecasts for Cimahi and Dewi Sartika with MAD values of 35.19 and 34.16, respectively. The selected forecasting results were subsequently used as input for transportation optimization. The initial transportation allocation obtained using the North West Corner (NWC) method generated a total cost of Rp.63,060,000. Further optimization using the Stepping Stone and Modified Distribution (MODI) methods successfully reduced the transportation cost to Rp.41,170,000, representing a significant improvement in distribution efficiency. The findings confirm that integrating forecasting and transportation models can improve demand estimation accuracy and minimize logistics costs simultaneously. This study contributes to the application of industrial engineering approaches in logistics and supply chain management, particularly for perishable product distribution systems.

Research on Inbound Logistics Demand Forecasting of Auto Parts Integrating Real-Time Production Plan Feedback and Error Compensation

Addressing nonlinear fluctuations in inbound logistics demand in the intelligent automotive industry, this paper proposes a SARIMA-LSTM-Attention forecasting model incorporating Real-Time Production Progress Feedback (RPF) features. By introducing production progress deviation into a hybrid framework that combines SARIMA-based linear forecasting with attention-enhanced LSTM residual correction, the proposed model effectively captures both linear trends and nonlinear demand disturbances. The model was evaluated using 995 daily observations collected from the inbound logistics system of a large new energy vehicle manufacturer from January 2023 to July 2025, with data from January 2023 to December 2024 used for training and validation and January to July 2025 reserved for testing. Compared with six representative benchmark models, the proposed model achieved the best overall performance, reducing sMAPE from 43.23% to 33.13% relative to the SARIMA baseline, representing an absolute reduction of 10.10 percentage points and a relative improvement of 23.36%. These results demonstrate the effectiveness of integrating real-time production feedback for demand forecasting and provide practical support for lean inventory management and logistics decision-making in automotive supply chains.

ARIMA & LSTM Model for Export Container Demand Forecasting to Support Sustainable Supply Chain Management