A Causal-XAI Framework for Retail Sales Forecasting: Addressing Transparency and Lack of Robustness in Deep Learning Models

Accurate sales forecasting in retail is very useful when managing inventory, planning replenishment strategy, and helping the business make good decisions in a retail environment. Various methods including Deep Learning (DL) and Ensemble techniques, have improved accuracy of sales forecasting in general [1] however, they do not distinguish between correlation and causal inference. Moreover, they cannot provide information on the underlying reasoning or rationale about the problem. Hence, these methods will produce a sales forecast and not provide actionable recommendations that would improve Return on Investment (ROI) related to pricing strategies or promotion strategies [2], [3]. Proposed new Causal-XAI framework guarantees reliable, trustworthy explanations for its predictions by using Double Machine Learning (DML) to correctly identify the true causes and Causal Transformer (CT) to use those causes for accurate and automatically clear predictions [4]. To apply our proposed framework, we are going to use Corporación Favorita Grocery Sales Forecasting dataset from Kaggle and evaluate the performance using the NWRMSLE [5]. The proposed method achieves not only superior forecasting accuracy but also presents causal information that directly leads to business actions aimed at improving marketing outcomes.