ASAP Hardware Failure-Cause Identification in Microwave Networks using Venn-Abers Pre

[simog72]

A good friend of mind recently published some interesting paper about AI in telco
https://re.public.polimi.it/retrieve...xtension-2.pdf

The study investigates how Machine Learning (ML) can be used to classify hardware failures in microwave networks. While ML methods are highly effective for this task, they typically provide predictions without indicating how reliable those predictions are—essentially, they don’t quantify the probability of being correct. In practice, gathering more data over longer timeframes can improve the accuracy of these models. However, this creates a challenge: balancing the need for high confidence in predictions with the desire to make those predictions as quickly as possible using minimal data.


To address this, the researcher reframes the problem of identifying hardware failure causes as an "As-Soon-As-Possible" (ASAP) classification challenge. In this approach, data arrives in a sequence, and the ML model provides a prediction only when its confidence surpasses a user-defined threshold. To achieve this, Inductive and Cross Venn-Abers Predictors are employed to enhance the probability estimates from any ML model, ensuring they are statistically reliable.


The results, based on real-world data, demonstrate that the ASAP framework can predict failures about 8 times faster than current methods, while maintaining over 95% accuracy for its selective classifications. Additionally, the dataset used in this research is publicly available to encourage further advancements in failure management for microwave networks.