Enhancing the product categories is crucial in the e-commerce industry since it directly affects operational effectiveness and profitability. Large and varied inventory is stored on e-commerce platforms, necessitating precise and effective category organization to improve customer experience and expedite processes. This study introduces AFLAT (Adaptive Feature Learning and Allocation Technique), a machine learning-based hierarchical method for overcoming the difficulties in classifying e-commerce products. To manage the intricate and ever-changing nature of consumer taxonomies, AFLAT employs a multi-layered architecture that integrates feature engineering, supervised learning, and adaptable categorization. The model uses a two-stage framework: the initial phase uses computer vision and natural language processing (NLP) techniques to collect and process essential information, such as written descriptions, images, and metadata. In order to successfully navigate the taxonomy, the second stage uses decision trees and deep learning models to construct a hierarchy of categorization mechanism that reflects the arrangement of e-commerce product categories. Seven machine learning algorithms—Multinomial Naive Bayes, Linear Support Vector Classifier, Multinomial Logistic Regression, Random Forest, XGBoost, FastText, and Voting Ensemble—were compared in order to achieve this aim. When compared to models that only use one strategy, this combined model performs better. According to the weighted F1-score, it exceeded the leading local approach (LocalClassifier per Level) model by 4.88% and overtook the best-performing flat method model by 0.15%. By providing a sizable Spanish-language dataset with more than a million products and going over the best preparation methods for the dataset, this paper also benefits the academic community. It also discusses possible directions for further research in this area as well as the study’s inherent limitations.