Explainability on the Fly for Vehicular Handover Management
A recent study published on arXiv (2608.14820) introduces a machine learning method that enhances explainability for managing handovers (HO) in vehicular networks. This research, categorized as a cross-type submission, tackles the issue of making rapid and dependable decisions in rapidly changing environments. While conventional ML models enhance HO detection by analyzing complex relationships among key performance indicators (KPIs), their opaque nature can hinder interpretability and trust from operators. To address this, the authors explore HO detection through an 'explainability-on-the-fly' approach, utilizing inherently interpretable models grounded in the functional analysis of variance (fANOVA) framework. The models are tested with two real-world operator datasets and benchmarked against a Long Short-Term Memory (LSTM) model with post-hoc SHAP explanations. In contrast to post-hoc methods, the fANOVA framework allows for immediate understanding of model decisions without extra computational costs, which is essential for latency-sensitive vehicular networks. The full paper can be accessed at https://arxiv.org/abs/2608.14820.
Key facts
- Paper arXiv:2608.14820 proposes explainable handover management for vehicular networks.
- Uses inherently interpretable models based on functional analysis of variance (fANOVA).
- Evaluated on two real-world operator datasets.
- Compared against LSTM baseline with post-hoc SHAP explanations.
- fANOVA framework provides immediate interpretation without extra computational overhead.
- Addresses black-box nature of ML in handover detection.
- Targets latency-sensitive vehicular networks.
- Available on arXiv with cross-type announcement.
Entities
Institutions
- arXiv