Decision-Centric Intelligent Marketing Analytics

In this paper, a Decision-Centric Intelligent Marketing An-alytics System (DCIMAS) for fashion e-commerce has been presented. The proposed system addresses three important research gaps: the gap between descriptive machine learn-ing predictions and prescriptive decision-making, the lack of uncertainty quantification for marketing managers, and the limited accessibility of AI-driven decision-support systems for small and medium-sized fashion businesses. The comparison of three supervised learning models on a fashion e-commerce dataset containing 500 customers with a positive-class ratio of 28.4% demonstrated that Logistic Re-gression achieved the highest F1-score (0.54) and recall (0.82). These results highlight the importance of selecting evaluation metrics appropriate for imbalanced binary classification. For the evaluated dataset, the interpretable linear model outper-formed the more complex ensemble approach in terms of F1-score and recall. Selecting a model based solely on accuracy would have favored Random Forest, potentially overlooking the stronger conversion-identification performance of Logistic Regression. The proposed confidence-based risk-scoring approach intro-duces uncertainty quantification into the marketing decision-making process. The confidence score, Ci = |Pi − 0.5| × 200, and the composite risk score Ri, which incorporates recency, RFM, and campaign response, provide actionable information for customer targeting. According to the experimental results, 24.2% of customers were classified as having low confidence and were flagged for manual review. This mechanism can help reduce the risk of making campaign decisions based on uncertain predictions near the model’s decision boundary. A key contribution of DCIMAS is its dual-layer recommen-dation approach. At the customer level, the recommendation engine generates fashion-oriented actions, including Premium Lookbook, Flash Sale, Win-back Offer, Discount Coupon, and Social Awareness Post, according to customer segments and recency. At the campaign level, the system assigns one of four operational tiers—Scale, Maintain, Optimize, or Pause—to each of the 50 campaigns based on its Weighted Sum Model (WSM) composite score. It also generates concise, human-readable recommendations incorporating campaign-specific ROI and conversion rate. The WSM-based campaign-ranking mechanism enables multicriteria evaluation using return on investment (ROI), conversion rate (CR), click-through rate (CTR), and cost per conversion (CpC). The evaluation identified five campaigns (10%) in the Scale tier and seven campaigns (14%) in the Pause tier, providing a basis for prioritizing budget allocation. The right-skewed distribution of campaign scores further in-dicates that a relatively small subset of campaigns achieved higher composite scores than the majority of the evaluated portfolio. Finally, the proposed open-source implementation uses Python, Flask, scikit-learn, and SQLite to provide an accessi-ble solution for fashion small and medium-sized businesses. Its CSV-based data workflow and portable file handling through Python’s pathlib reduce deployment complexity and elimi-nate the need for cloud infrastructure. Overall, DCIMAS inte-grates predictive analytics, uncertainty-aware customer assess-ment, campaign ranking, and actionable recommendations into a unified decision-support framework for fashion e-commerce marketing.

[1]     “Research on the Application of Data Mining Technology in Enterprise Marketing,” ACM, 2024. [2]     “The Role of Big Data in Modern Marketing Decisions,” IEEE, 2024. [3]     “Machine Learning-Based Market Segmentation and Consumer Behav-ior Prediction Models,” ACM, 2024. [4]     “Data Mining and Intelligent Decision-Making in Digital Accounting Information Management Systems,” ACM, 2024. [5]     “Machine Learning-Driven Consumer Behavior Analysis: An AI Ap-proach for Intelligent Marketing,” IEEE, 2025. [6]     “Enhancing Customer Repurchase Prediction: Integrating Classification Algorithms with RFM Analysis for Precision and Actionable Insights,” ScienceDirect, 2025. [7]     “An Adaptive Analytics Framework for Customer Retention Through In-tegrative Feature Optimization and Ensemble Learning,” ScienceDirect, 2025. [8]     “An Effective Machine Learning Model for Accurate ROI Prediction and Optimization in Marketing Strategies,” IEEE, 2025. [9]     “Intelligent Analysis Methods for Multi-channel Marketing Data Based on Anomaly Detection Algorithms,” ACM, 2025. [10]    “AI-Driven Marketing Strategy Optimization in the Digital Economy: A Machine Learning Approach,” ACM, 2025.