Analyses of automated machine learning tools for application in marketing

• Veröffentlicht in: PROBLEMS IN PROGRAMMING
• Datum: 2026
• Heft: 1
• Сторінки: 93-101
• ISSN: 1727-4907
• Автори та афіліації: O.V. Nikonov — NTUU "KPI"
• Ключові слова: предиктивна аналітика, прогнозування відтоку клієнтів, роздрібна торгівля, прогнозування продажів, автоматизоване машинне навчання, прогнозна аналітика, машинне навчання, маркетингова аналітика, модель машинного навчання, інженерія ознак
• 1. Verfasser: Nikonov, O.V.
• Format: Artikel
• Sprache: Ukrainisch
• Veröffentlicht: PROBLEMS IN PROGRAMMING 2026
• Schlagworte: automated machine learning; predictive analytics; customer churn prediction; retail; feature engineering; UDC 004.052.32; 004.43
• Online Zugang: https://pp.isofts.kiev.ua/index.php/ojs1/article/view/894
• Назва журналу: Problems in programming

The article investigates the problem of automating the activities of IT experts in machine learning using modern AutoML frameworks (AutoSklearn and TPOT). The aim of the work is to overcome the fundamental contradiction between the high resource intensity of manual creation of predictive pipe lines and the need to build accurate models in conditions of limited data, when companies have only basic transactional information. The proposed approach formalizes the use of AutoML algorithms to solve the problem of predicting customer churn in retail, replacing manual data processing processes with automated solutions. The approach is implemented by generating a set of subject-oriented features based on the RFM model and validated by historical simulation methods on the transactional dataset "Online Retail". Experimental results demonstrate that AutoML systems are able to work effectively with "raw" data: AutoSklearn provides a stable weighted F1-measure at the level of 0.78 and ROC AUC 0.792 in just 5 minutes of work. The work has practical significance for developing resource efficient predictive systems, minimizing the impact of the human factor, and accelerating the deploy ment of models at enterprises with a basic level of data collection.Problems in programming 2026; 1: 93-101