MACHINE LEARNING IMPLEMENTATION IN LAST-MILE INBOUND PARCEL VOLUME PREDICTION FOR WORKFORCE SCHEDULING IMPROVEMENT IN A LOGISTICS COMPANY

• Main Author: Richard
• Format: Theses
• Language: Indonesia
• Subjects: Manajemen umum
• Online Access: https://digilib.itb.ac.id/gdl/view/86717
• Institution: Institut Teknologi Bandung

The last-mile delivery sector in logistics faces challenges like fluctuating demand and unpredictable delivery volumes, while requiring efficient workforce scheduling to optimize costs, service quality, and customer satisfaction. This study investigates the use of time series forecasting techniques to improve workforce scheduling in a logistics company’s last-mile delivery operations. The research explores forecasting models such as Holt-Winters Exponential Smoothing, Seasonal ARIMA, and DeepAR, applied to historical delivery data to predict demand and optimize workforce allocation. These models address complexities like seasonal fluctuations, long-term trends, and irregular demand spikes typical in last-mile logistics. A case study using time-stamped delivery records, order volumes, and geographical data reveals that Holt-Winters captures seasonal demand, while Seasonal ARIMA provides more robust long-term forecasts. DeepAR, a machine learning model, captures intricate patterns and outperforms others in handling complex seasonal dynamics. This dynamic approach adjusts staffing based on predicted demand, reducing overstaffing and ensuring adequate resources during peak periods, improving service reliability. In conclusion, the study demonstrates that time series forecasting enhances workforce scheduling accuracy and improves worker satisfaction in last-mile delivery. Future research could explore incorporating external factors and advanced machine learning techniques to further enhance forecasting accuracy.