The rapid growth of software vulnerabilities demands advanced prioritization beyond static scoring systems. Recent works (2020-2025) have applied machine learning (ML) and large language models (LLMs) to predict and rank the risk of vulnerabilities based on features such as CVSS metrics, exploit presence, context and natural language descriptions. This review surveys supervised, unsupervised, and hybrid ML approaches-including neural networks, ensemble classifiers, graph-based models-and LLM-based NLP methods. We examine various model types (e.g. Random Forest, XGBoost, CNN, DistilBERT), data sources (NVD/CVE descriptions, exploit databases, OSINT, telemetry), evaluation metrics (accuracy, F1-score, MSE), and quantitative results. Performance trends indicate ML can improve on CVSS baselines (e.g.$\boldsymbol{\sim} \mathbf{8 3 \%}$accuracy), whereas LLMs require domain adaptation to perform well. Finally, we discuss gaps in research, such as limited real-world validation and the need for dynamic, context-aware models.

Propulsion-fault anticipation is vital for multirotor-UAV safety. We formulate motor-fault prediction as a 3s earlywarning binary task and evaluate three lightweight classifiers–Logistic Regression (LR), Random Forest (RF) and Gradient Boosting (GB)–on a curated subset of the public RflyMAD corpus (34252 sliding windows, 8.1 % faults). Each window is summarised by 64 statistical features (mean, standard deviation, minimum, maximum) derived from 16 telemetry channels and standardised on the training split only. A leak-free GroupKFold ($k=3$) protocol ensures flight-wise separation; class imbalance is corrected with RandomOverSampler, and a 15 % validation slice sets the F1-optimal threshold. On the outer test folds RF delivers the best compromise, achieving$\mathrm{F}_{1}=0.850 \pm 0.003$, ROC-AUC$=0.994 \pm 0.000$and a false-alarm rate (FAR)$=21 \pm 14 \mathbf{h}^{-\mathbf{1}}$, while issuing alerts$3.67 \pm 2.88 ~\mathrm{s}$before failure. GB follows closely$\left(F_{1}=0.832; F A R=25 ~\mathrm{h}^{-1}\right)$, whereas LR trails ($F_{1} =0.649; \text{FAR}=64 \mathbf{h}^{-\mathbf{1}}$). All three models satisfy the 3 -s warning criterion, but tree-based ensembles offer a superior recall-to-false-alarm balance and sub-millisecond inference, meeting edge-deployment constraints without deep-learning complexity. These findings support the use of certifiable classical models for on-board UAV fault prognostics and motivate future validation on real flights and multi-fault scenarios.