Alexander Bakumenko

Mr. Alexander Bakumenko

Luleå University of Technology, Luleå, Norrbotten, Sweden
+1
orcid

Research Focus & Metrics

Academic Career Length

5Years (2022-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
25 August 2022 in Systems
Verified
Journal Article
Open access
PDF file
Alexander BakumenkoAlexander Bakumenko
Alexander Bakumenko
Ahmed ElragalAhmed Elragal
Ahmed Elragal
Bookkeeping data free of fraud and errors are a cornerstone of legitimate business operations. The highly complex and laborious work of financial auditors calls for finding new solutions and algorithms to ensure the correctness of financial statements. Both supervised and unsupervised machine learning (ML) techniques nowadays are being successfully applied to detect fraud and anomalies in data. In accounting, it is a long-established problem to detect financial misstatements deemed anomalous in general ledger (GL) data. Currently, widely used techniques such as random sampling and manual assessment of bookkeeping rules become challenging and unreliable due to increasing data volumes and unknown fraudulent patterns. To address the sampling risk and financial audit inefficiency, we applied seven supervised ML techniques inclusive of deep learning and two unsupervised ML techniques such as isolation forest and autoencoders. We trained and evaluated our models on a real-life GL dataset and used data vectorization to resolve journal entry size variability. The evaluation results showed that the best trained supervised and unsupervised models have high potential in detecting predefined anomaly types as well as in efficiently sampling data to discern higher-risk journal entries. Based on our findings, we discussed possible practical implications of the resulting solutions in the accounting and auditing contexts.

20 August 2025 in IEEE Access
Verified
Journal Article
Open access
Alexander BakumenkoAlexander Bakumenko
Alexander Bakumenko
Kateřina Hlaváčková-SchindlerKateřina Hlaváčková-Schindler
Kateřina Hlaváčková-Schindler
Claudia PlantClaudia Plant
Claudia Plant
NHNina C. Hubig
Nina C. Hubig
Detecting anomalies in general ledger data is of utmost importance to ensure the trustworthiness of financial records. Financial audits increasingly rely on machine learning (ML) algorithms to identify irregular or potentially fraudulent journal entries, each characterized by a varying number of transactions. In machine learning, heterogeneity in feature dimensions adds significant complexity to data analysis. In this paper, we introduce a novel approach to anomaly detection in financial data using Large Language Model (LLM) embeddings. To encode non-semantic categorical data (i.e., attributes lacking inherent linguistic meaning) from real-world financial records, we tested 3 pretrained general-purpose sentence-transformer models. For the downstream classification task, we implemented and evaluated 5 optimized ML models, including Logistic Regression, Random Forest, Gradient Boosting Machines, Support Vector Machines, and Neural Networks. Our experiments demonstrate that LLMs contribute valuable information to anomaly detection as our models outperform the baselines, in selected settings by a large margin. The findings further underscore the effectiveness of LLMs in enhancing anomaly detection in financial journal entries, particularly by tackling feature sparsity. We discuss a promising perspective on using SBERT embeddings for non-semantic data in the financial context and beyond.

26 March 2026 in Research Square
Verified
Preprint
Open access
Alexander BakumenkoAlexander Bakumenko
Alexander Bakumenko
JHJanine Hoelscher
Janine Hoelscher
DSD. Hudson Smith
D. Hudson Smith
SBSvetlana Bakumenko
Svetlana Bakumenko
AMAaron J. Masino
Aaron J. Masino
JOJihad S. Obeid
Jihad S. Obeid
Background : Multimodal machine learning for intensive care unit (ICU) in-hospital mortality (IHM) prediction has rapidly expanded, yet reporting practices relevant to clinical readiness, including calibration, clinical utility, robustness to data missingness, and actionable transparency, remain heterogeneous. Objective : To map the landscape of multimodal IHM prediction studies and synthesize evidence on (i) fusion architectures and transparency implementations, (ii) evaluation beyond discrimination (calibration, clinical utility, and per-case uncertainty), and (iii) operational readiness (missing modality handling, prediction timeliness, external validation, subgroup evaluation, and reproducibility artifacts). Methods : We conducted a scoping review of peer-reviewed studies (English, January 2010 – November 2025) that fused at least two data modalities by provenance and predicted IHM in adult ICU populations. Searches were performed in PubMed, Scopus, and Google Scholar. Screening and extraction used an established AI-assisted workflow with two-reviewer verification. We summarized extracted fields from 115 included studies using descriptive statistics and narrative synthesis, including thematic analysis of author-reported gaps, limitations, and future directions. Results : Among 115 included studies, feature-level (early) fusion predominated (n = 80, 69.6%), followed by intermediate fusion (n = 30, 26.1%), late fusion (n = 4, 3.5%), and hybrid fusion (n=1, 0.9%). The modality landscape was dominated by physiologic time-series and coded/tabular EHR data; clinical text, imaging, and high-frequency waveforms each appeared in a minority of studies. Feature-level interpretability was reported in 21 studies (18.3%), with full per-case attribution in only 13 (11.3%); modality-level attribution was identified in 2 studies (1.7%), and missing-modality policies in 7 (6.1%). Calibration was assessed in 40 studies (34.8%), decision curve analysis in 20 (17.4%), and per-case uncertainty quantification in 1 (0.9%). External validation was performed in 28 studies (24.3%), prediction timeliness was evaluated in 22 (19.1%), and code repositories were provided in 29 (25.2%); no study reported a structured model card. Conclusions : The multimodal IHM literature emphasizes discrimination performance and early fusion, yet the evidence base remains immature for clinical translation, with systematic gaps in calibration, clinical utility, per-case transparency, missing-modality robustness, and reproducibility. These findings motivate standardized reporting and evaluation frameworks tailored to multimodal ICU prediction.

19 November 2025 in arXiv
Verified
Preprint
Open access
PDF file
Alexander BakumenkoAlexander Bakumenko
Alexander Bakumenko
JHJanine Hoelscher
Janine Hoelscher
HSHudson Smith
Hudson Smith
Early identification of intensive care patients at risk of in-hospital mortality enables timely intervention and efficient resource allocation. Despite high predictive performance, existing machine learning approaches lack transparency and robustness, limiting clinical adoption. We present a lightweight, transparent multimodal ensemble that fuses physiological time-series measurements with unstructured clinical notes from the first 48 hours of an ICU stay. A logistic regression model combines predictions from two modality-specific models: a bidirectional LSTM for vitals and a finetuned ClinicalModernBERT transformer for notes. This traceable architecture allows for multilevel interpretability: feature attributions within each modality and direct per-case modality attributions quantifying how vitals and notes influence each decision. On the MIMIC-III benchmark, our late-fusion ensemble improves discrimination over the best single model (AUPRC 0.565 vs. 0.526; AUROC 0.891 vs. 0.876) while maintaining well-calibrated predictions. The system remains robust through a calibrated fallback when a modality is missing. These results demonstrate competitive performance with reliable, auditable risk estimates and transparent, predictable operation, which together are crucial for clinical use.

21 May 2026 in medRxiv
Verified
Preprint
Open access
PDF file
Alexander BakumenkoAlexander Bakumenko
Alexander Bakumenko
DSD. Hudson Smith
D. Hudson Smith
JHJanine Hoelscher
Janine Hoelscher
Earlier ICU mortality prediction is more clinically useful because it can identify high-risk patients while treatment decisions can still change. Yet most models are trained on data from a fixed time window, so it is unclear whether a model trained on the first 48 hours of ICU data remains reliable when used earlier in the ICU stay. We evaluated a multimodal ICU mortality model trained once at 48 hours and then applied unchanged at 6, 12, 24, and 48 hours on MIMIC-III. The model combines an LSTM for physiological time-series data, a finetuned ClinicalModernBERT model for clinical notes, and a logistic regression fusion layer. Performance remained strong at earlier time points, suggesting that useful mortality prediction is possible earlier in the ICU stay even without retraining. At 6 hours, the model achieved AUROC 0.777 and remained well-calibrated (expected calibration error, ECE 0.038) without any recalibration, and it outperformed both single-modality models at every horizon. The benefit of combining both modalities was most evident at earlier horizons, when physiological data were sparse: agreement between the two specialists dropped by more than half from 48 to 6 hours, while the median contribution from clinical notes increased from 37% to 49%. A Bayesian version of the fusion layer showed that uncertainty decreased for survivors as more data accumulated but remained high for non-survivors; the most uncertain cases were up to 4.9 times more likely to be non-surviving patients. Continuous hourly analyses further showed that clinical notes provide stable context between documentation events. Simply carrying forward the most recent note matched or outperformed note-decay and documentation-gap alternatives. These results suggest that a multimodal ICU mortality model trained on 48 hours of data can provide trustworthy earlier predictions without retraining, while also identifying the cases that remain hardest to interpret.

31 October 2025 in arXiv
Verified
Preprint
Open access
PDF file
Alexander BakumenkoAlexander Bakumenko
Alexander Bakumenko
AMAaron J. Masino
Aaron J. Masino
JHJanine Hoelscher
Janine Hoelscher
The translation of artificial intelligence (AI) systems into clinical practice requires bridging fundamental gaps between explainable AI theory, clinician expectations, and governance requirements. While conceptual frameworks define what constitutes explainable AI (XAI) and qualitative studies identify clinician needs, little practical guidance exists for development teams to prepare AI systems prior to clinical evaluation. We propose two foundational design principles, Transparent Design and Operable Design, that operationalize pre-clinical technical requirements for healthcare AI. Transparent Design encompasses interpretability and understandability artifacts that enable case-level reasoning and system traceability. Operable Design encompasses calibration, uncertainty, and robustness to ensure reliable, predictable system behavior under real-world conditions. We ground these principles in established XAI frameworks, map them to documented clinician needs, and demonstrate their alignment with emerging governance requirements. This pre-clinical playbook provides actionable guidance for development teams, accelerates the path to clinical evaluation, and establishes a shared vocabulary bridging AI researchers, healthcare practitioners, and regulatory stakeholders. By explicitly scoping what can be built and verified before clinical deployment, we aim to reduce friction in clinical AI translation while remaining cautious about what constitutes validated, deployed explainability.