Pharmaceutical Sales Forecasting using Machine Learning

Accurate pharmaceutical sales forecasting is crucial for managing inventory, improving supply chains, and lowering financial risks from stockouts and product expiries. Traditional statistical methods like ARIMA often struggle to address nonlinear dependencies and irregular demand patterns found in real retail environments. In response, recent research has increasingly used machine learning approaches that show better accuracy and
flexibility. This paper reviews a collection of recent studies on time series forecasting, focusing on methods for data preprocessing, feature engineering, and model development. Based on the findings from these studies, the paper presents a structured forecasting perspective that combines effective preprocessing strategies with
machine learning techniques suited for diverse pharmaceutical datasets. Special emphasis is placed on tree-based ensemble models like XGBoost for managing structured retail data and neural network methods for situations with limited historical records. The discussion highlights how these complementary techniques can
work together to tackle challenges such as demand fluctuations, sparse data conditions, and support for operational decisions in pharmaceutical supply chains. Comparative results from the studies underscore the reliability of XGBoost in handling structured datasets and the performance of GRNN in low-data scenarios. The
discussion also addresses key limitations, such as interpretability and scalability, and suggests future directions for real-world application.