Juan C. Alfaro

Research profile

Research summary

Juan C. Alfaro is a machine learning researcher at the University of Castilla-La Mancha (Spain), profiled from public OpenAlex bibliographic records (ORCID 0000-0003-1777-8540). His recorded output spans 2018–2026, with per-year counts rising from one work in 2018–2019 to four in each of 2025 and 2026, and 10 works in the last three years. The snapshot reports an h-index of 5 and an i10-index of 1, while a validated citation-impact score is not yet computed. His work clusters around data classification, with three equally weighted strands (0.333 each): machine learning, federated learning with Bayesian classifiers, and time-series anomaly detection for spacecraft telemetry. Three 2026 arXiv preprints reviewed by Evolve illustrate this range. As corresponding author of MORE-PLR, he applies multi-output regression to partial label ranking; Evolve judged the meta-framework sound and competitive with state-of-the-art ranking-by-classification at lower cost. FedAnDE is described as the first federated framework for discriminative Averaged n-Dependence Estimators (n≥1), with gains over differentially private generative naive Bayes. A third preprint applies EVT-thresholded unsupervised anomaly detection to the ESA-AD spacecraft telemetry benchmark (F0.5 0.700/0.698). Evolve's reproducibility review averages 70/100 across the three papers, one rated high and two medium readiness.

Research keywords

Highlights

• MORE-PLR: multi-output regression for partial label ranking

  Corresponding-author 2026 arXiv preprint reformulating partial label ranking as multi-output regression. Evolve's verdict: a sound and competitive alternative to state-of-the-art ranking-by-classification at lower cost, with benchmark-based tuning and a reporting anomaly flagged. Reproducibility readiness: high.

• FedAnDE: first federated framework for discriminative AnDE

  2026 arXiv preprint described by Evolve as the first federated framework for discriminative Averaged n-Dependence Estimators (n≥1), showing solid empirical gains over differentially private generative naive Bayes, while its privacy claims lack formal backing. Reproducibility readiness: medium.

• Label-free anomaly detection for spacecraft telemetry

  Structure-aware unsupervised detection with adaptive EVT thresholding. Evolve finds the reported ESA-AD F0.5 of 0.700/0.698 credible, but notes that model choice used full-mission data and that M1 recall is 0.408. Reproducibility readiness: medium.

Timeline

1. 2018Earliest recorded year in the per-year work counts, with 1 work; the count stays at 1 in 2019.
2. 2021Early steady output: 2 works recorded in each of 2020 and 2021, within the machine learning and data classification topic cluster.
3. 20233 works recorded in 2023, followed by 2 in 2024, per the per-year counts.
4. 2025Activity accelerates to 4 works — the highest annual count recorded up to that point.
5. 20264 works recorded; three arXiv preprints enter the Evolve catalog — MORE-PLR (corresponding author), FedAnDE, and spacecraft-telemetry anomaly detection — with reproducibility averaging 70/100.

Papers in the Evolve catalog

• 2026-09-24corresponding

  MORE-PLR: multi-output regression employed for partial label ranking

• 2026-09-23

  Federated Learning of AnDE Classifiers

• 2026-09-09

  Structure-Aware Unsupervised Anomaly Detection for Spacecraft Telemetry with Adaptive EVT Thresholding

Frequent co-authors