New Feature-Selection Algorithm Predicts Aortic Calcification in Dialysis Patients

Because its development is shaped by many clinical features and metabolic indicators, clinicians need accurate ways to predict who will develop it early enough to intervene. A new study published in Applied Intelligence reports a machine-learning pipeline designed for exactly this task, combining an enhanced optimization algorithm with a support vector machine classifier.

The core of the method is an improved version of the kernel search optimizer, a metaheuristic used to select the most informative features from a larger set. Redundant features can degrade predictive accuracy, so the researchers introduced two modifications. An adaptive topology strategy combines information from the current individual and its neighbors to capture differences between regions of the search space, thereby increasing population diversity. A competitive cooperation strategy adds intra-population competition and inter-population communication to strengthen the algorithm’s local exploration capability. The resulting optimizer, ACKSO, is converted to a binary form, bACKSO, which searches for the optimal feature subset before a support vector machine predicts abdominal aortic calcification from the selected features.

According to the reported results, the bACKSO-SVM approach achieved an accuracy of 92.8 percent, a sensitivity of 92.7 percent, and a specificity of 93.4 percent, along with a precision of 93.2 percent, an MCC of 0.858, an F-measure of 0.928, and an AUC of 0.907. These figures indicate that the model balanced correctly identifying patients with calcification against correctly clearing those without it, a balance that matters clinically because both missed cases and false alarms carry costs in a dialysis population already at elevated cardiovascular risk.

Beyond raw performance, the feature-selection stage yielded a compact set of five key features strongly associated with abdominal aortic calcification prediction: age, statin use, the 3-hour intra-dialytic diastolic blood pressure reading, serum calcium, and dialysis access dysfunction. That list spans demographics, medication, a measurement taken mid-dialysis, a metabolic indicator, and a vascular-access complication, illustrating how the algorithm distilled a heterogeneous clinical record into a small, interpretable panel.

The authors frame the work as a step toward early prediction that could enable earlier intervention in a population where calcification is prevalent and consequential. As with any model reported at this stage, the figures come from the study’s own evaluation of the proposed method, and the abstract describes promising predictive performance rather than deployed clinical use. Still, the combination of a diversified search strategy, a competitive cooperation mechanism, and a classifier tuned to a reduced feature set offers a concrete template for prediction tasks where many candidate clinical variables compete for a place in the final model.

Subject of Research: Machine-learning feature selection for early prediction of abdominal aortic calcification in maintenance hemodialysis patients

Article Title: Adaptive and cooperative kernel search optimizer-based feature selection for early prediction of abdominal aortic calcification

Article References: Chen, S., Cao, Q., Wu, Y., Xu, L., Cai, Z., Chen, H., & Bu, S. (2026). Adaptive and cooperative kernel search optimizer-based feature selection for early prediction of abdominal aortic calcification. Applied Intelligence, 56(16), Article 527. https://doi.org/10.1007/s10489-026-07511-8

Image Credits: AI Generated

DOI: 10.1007/s10489-026-07511-8

Keywords: abdominal aortic calcification, hemodialysis, feature selection, kernel search optimizer, support vector machine, cardiovascular mortality, machine learning, serum calcium, blood pressure, predictive model, Adaptive, cooperative