Automation of modeling of business solution development scenarios

DOI: 10.31673/2412-9070.2026.318115

Abstract

The article is devoted to the automation of modeling scenarios for the development of business solutions based on the analysis of behavioral metrics of web project users. The study is based on a test sample of data that simulates real indicators from analytical systems (Google Analytics, Bino-tel, CRM), including parameters: region (city), number of page views (page_view), clicks on advertising (ad_click), fact of purchase (purchase) and duration of interaction (duration). The full sample covers 20 records from different regions of Ukraine, which allows modeling digital interaction. The first stage is a correlation analysis of the dependence between page_view and ad_click. The Shapiro-Wilk normality test showed a deviation from normality (p < 0.05), therefore, the Spearman coefficient was applied (ρ = 0.8548, p < 0.001), which confirms a strong direct relationship. The analysis was implemented in Python using pandas, matplotlib, seaborn, and scipy libraries, with visualization as a scatter plot. The DecisionTreeClassifier decision tree classification model was applied to predict purchases. The model achieved 95% accuracy, with metrics precision 0.91–1.00, recall 0.90–1.00, f1-score 0.95. The decision tree was visualized, and borderline error cases were identified (for example, for Dnipro). The model logic includes conditions based on page_view and ad_click, which provides interpretability for integration into business logic. The next step is logistic regression (Logis-ticRegression) to estimate the probability of purchase. Accuracy 85%, 17 out of 20 objects were correctly classified; error matrix, metrics: precision 0.82–0.89, recall 0.80–0.90, f1-score 0.84–0.86. The classification boundary and logistic curves for the purchase ~ page_view and purchase ~ ad_click dependencies are visualized, demonstrating a positive relationship. Cox regression (proportional hazards model) was used to account for the time factor. The ad_click feature has a positive effect (log(HR) ≈ 1.1, p=0.19), page_view has a weak negative effect (log(HR) ≈ -0.08, p=0.44), but without statistical significance due to the small sample. Data expansion is recommended to increase accuracy. In a practical example for an e-commerce company with low conversion (1.8%), an intelligent system with LDA and QDA (100% accuracy) was modeled. Scenario modeling of actions (e.g., add_product_reviews, personalize_content, optimize_landing_page) increased the predicted conversion to 50–70%, achieving the target of 2.5% with minimal costs. The system integra-tes models for adaptive business strategies. The proposed approach automates scenario modeling, combining statistical methods with machine learning to optimize business decisions in a digital environment. The results confirm the effectiveness for predicting behavior and shaping actions, with the potential for scalability.

Keywords: business decisions, behavioral metrics, correlation analysis, decision tree, logistic regression, Cox regression, machine learning, conversion, automation.