D.S. Ripka1, E.A. Semenova2

1, 2 St. Petersburg State Electrotechnical University "LETI" (St. Petersburg, Russia)
1 dar.stkr@gmail.com, 2easemenova@etu.ru

The development of effective assisted reproductive technologies (ART) is complicated by limited clinical samples, high dimensionality of diagnostic feature space, and low interpretability of machine learning models, which hinders their implementation into clinical practice. Existing statistical and data-driven approaches do not allow fully formalizing and utilizing the expert knowledge of reproductive medicine specialists accumulated over decades of practical work.

The aim of the study is to develop a methodology for hybrid weighting of diagnostically significant indicators based on quantitative expert consensus with the construction of an interpretable decision tree for predicting ART effectiveness.

A scientific substantiation of the hybrid approach to weighting diagnostically significant indicators is carried out. Based on a survey of experts and two-stage outlier filtration (IQR + modified Z-method), normalized weights for 43 diagnostically significant indicators are obtained. Indicators with high expert consensus (coefficient of variation <10%), such as medical errors and ovulation level, receive maximum weights (0.97–0.98). An additive predictive decision tree model with logistic calibration is constructed, allowing the transformation of diagnostic data into a calibrated probability of successful ART outcome. The model provides a monotonic increase in prediction accuracy from 55% (in the absence of data) to 100% (with a full set of features).

The use of developed methodology is applicable in clinical decision support systems even with small clinical samples, since feature weights are determined a priori by experts and do not require training on large datasets. The interpretable structure of the decision tree allows the physician to track the contribution of each diagnostic indicator to the final prediction, which increases trust in the system's recommendations and contributes to personalized patient management in ART programs.

Ripka D.S. Hybrid expert model for predicting ART outcomes based on weighted decision tree and logistic calibration. Biomedicine Radioengineering. 2026. V. 29. № 3. P. 119–124. DOI: https:// doi.org/10.18127/ j15604136-202603-21 (In Rus-sian)

1. Bruckamp L., Lazzari E. Shifting the reproductive window: The contribution of ART and egg donation to fertility rates in the UK. Popul Stud (Camb). 2025. Oct 27. P. 1–13. DOI: 10.1080/00324728.2025.2561595.
2. Shanti S. Gupta. Percentage Points and Modes of Order Statistics from the Normal Distribution. The Annals of Mathematical Statistics. Sep., 1961. V. 32. № 3. P. 888–893.
3. Christophe Leys et al. Detecting outliers: Do not use standard deviation around the mean, use absolute deviation around the median. Journal of Experimental Social Psychology. 2013. V. 49, Is. 4. P. 764–766. https://doi.org/10.1016/j.jesp.2013.03.013.
4. Kerrigan D., Hullman J., Bertini E. A Survey of Domain Knowledge Elicitation in Applied Machine Learning. Multimodal Technologies and Interaction. 2021. V. 5(12). P. 73. https://doi.org/10.3390/mti5120073.
5. Sarker I.H., Kayes A.S.M. Watters P. Effectiveness analysis of machine learning classification models for predicting personalized context-aware smartphone usage. J. Big Data. 2019. V. 6. P. 57. https://doi.org/10.1186/s40537-019-0219-y.
6. Topol E.J. High-performance medicine: the convergence of human and artificial intelligence. Nat Med 25. 2019. P. 44–56. https://doi.org/10.1038/s41591-018-0300-7.
7. Semenova E., Ripka D. Formation of a Complex of Diagnostically Significant Indicators for Predicting the Outcomes of Assisted Reproductive Technologies and Pregnancy Pathologies. STDH. 2025. P. 294–296. 10.1109/STDH66836.2025.11227529.