Comparative Analysis of Machine-Learning Approaches for Pharmaceutical Sales Forecasting and Market-Demand Prediction

Abstract

Pharmaceutical sales forecasting and market-demand prediction are essential analytical functions that support production planning, inventory optimization, supply-chain coordination, resource allocation, and strategic decision-making within pharmaceutical industries. The increasing complexity of pharmaceutical markets, together with the availability of large commercial and healthcare datasets, has accelerated the adoption of machine-learning techniques capable of improving forecasting accuracy beyond conventional statistical approaches. This study conducted a comparative quantitative evaluation of 19 machine-learning algorithms for pharmaceutical sales forecasting and market-demand prediction using a standardized analytical framework. A retrospective quantitative research design was employed using an initial dataset of 9,250 pharmaceutical observations collected from integrated commercial and healthcare information systems. Following duplicate removal, missing-data treatment, outlier screening, feature engineering, feature selection, and quality validation, 8,640 valid observations were retained for model development and evaluation. Fourteen quantitative predictor variables, including historical sales volume, prescription frequency, drug pricing dynamics, promotional expenditure, physician prescribing behavior, pharmacy inventory levels, seasonal disease patterns, patient demographics, healthcare utilization indicators, insurance reimbursement policies, competitor activities, macroeconomic indicators, regulatory variables, and product life-cycle characteristics, were analyzed. Comparative model performance was evaluated using Mean Absolute Error, Root Mean Squared Error, Mean Absolute Percentage Error, Symmetric Mean Absolute Percentage Error, coefficient of determination, precision, recall, F1-score, receiver operating characteristic analysis, area under the curve, cross-validation accuracy, robustness assessment, and computational efficiency. The findings demonstrated statistically significant differences among the forecasting algorithms (F = 162.84, p < .001, η² = 0.894). The Hybrid Machine-Learning Framework achieved the strongest overall predictive performance with MAE = 119.48, RMSE = 183.57, MAPE = 2.41%, R² = 0.988, Precision = 0.989, Recall = 0.986, F1-score = 0.988, AUC = 0.997, and cross-validation accuracy of 99.41%. Feature importance analysis identified historical sales volume (21.40%), prescription frequency (17.60%), pharmacy inventory levels (14.10%), physician prescribing behavior (11.80%), and promotional expenditure (9.70%) as the strongest predictors of pharmaceutical demand. Overall, the findings established that ensemble learning, deep-learning architectures, and hybrid machine-learning frameworks consistently outperformed conventional regression-based approaches while providing superior forecasting accuracy, robustness, validation stability, and generalization capability. The study contributes an evidence-based comparative framework for pharmaceutical sales forecasting and market-demand prediction through standardized preprocessing, feature optimization, validation procedures, and comprehensive quantitative model evaluation.