Contemporary retail operations are confronted with escalating complexity in demand forecasting and inventory management, driven by non-linear demand variability, promotional amplification, seasonal fluctuations, and the competitive imperatives of omnichannel commerce. Conventional forecasting paradigms-predicated upon moving averages, exponential smoothing, and linear extrapolation-demonstrably fail to capture the multivariate, nonlinear interactions governing retail demand dynamics. This paper introduces SmartRetail AI, a comprehensive, end-to-end retail intelligence platform that integrates ensemble machine learning with classical inventory optimisation theory within a unified data architecture. The system is constructed upon a structured synthetic dataset encompassing 4,250 daily transaction records across five representative Fast-Moving Consumer Goods and e-commerce product categories, spanning 850 operational days. The proposed forecasting engine deploys productlevel Random Forest Regressor models-each trained independently to capture category-specific demand dynamics-to generate thirty-day forward demand forecasts. Engineered temporal features including day-of-week indicators, monthly seasonality encodings, weekend binary flags, and promotional activity markers constitute the input feature space. The inventory optimisation module applies the classical safety stock formulation at a 95% service level target (Z = 1.65), computing product-specific reorder points as a function of historical demand variability and a three-day supply lead time. Analytical outputs are persisted in a normalised MySQL relational schema comprising three tables-historical sales, demand forecasts, and inventory metrics-and rendered through an executive intelligence dashboard. Comparative evaluation against moving average baselines demonstrates Mean Absolute Error reductions of 23.4% to 47.8% across product categories, with Root Mean Square Error improvements ranging from 19.7% to 44.1%. The platform achieves a Mean Absolute Percentage Error of 12.7% for high-volume staples and 24.3% for low-velocity electronics, establishing its operational viability across diverse retail demand profiles. These results validate the proposed framework as a practically deployable, analytically rigorous alternative to legacy forecasting methodologies.
Building similarity graph...
Analyzing shared references across papers
Loading...
Gajula Sai Rohith
V. Vijayalakshmi
Building similarity graph...
Analyzing shared references across papers
Loading...
Rohith et al. (Thu,) studied this question.
www.synapsesocial.com/papers/69ec593e88ba6daa22dab42e — DOI: https://doi.org/10.64672/ijifr/26.04.13.08.051