Predictive Mathematical Modeling and Classification of Retail Sales Orders Using AI Machine Learning Techniques

Abstract

This study presents a systematic mathematical model known as a low-code approach to classifying retail sales orders by size using AI machine learning techniques within the Orange Data Mining platform. Leveraging a real-world sales dataset sourced from Kaggle, we implemented and evaluated ten classification models, including ensemble learners (AdaBoost, Gradient Boosting), probabilistic classifiers (Naïve Bayes), distance-based methods (kNN), and interpretable algorithms (CN2 Rule Induction). Each model was assessed through 10-fold cross-validation using performance metrics such as accuracy, F1-score, precision, recall, AUC, and LogLoss. The experimental workflow integrated visual preprocessing, model training, and comparative evaluation, enabling reproducibility without programming expertise. The results reveal that ensemble models, particularly AdaBoost, achieved perfect classification accuracy (100%) and AUC (1.000), while CN2 Rule Induction offered near-perfect accuracy (99.8%) alongside interpretable rule-based outputs. Traditional models like Logistic Regression and kNN also demonstrated strong performance but were outperformed by advanced ensembles. This research contributes a novel combination of high-performing and explainable models in a retail classification task using a low-code framework. The proposed approach provides practical guidance for retailers, analysts, and educators seeking accurate and accessible predictive tools for operational decision-making. Future directions include multi-class extension, imbalance handling, and deployment in real-time environments.