Particle detectors such as resistive plate chambers (RPCs) work like sensitive traps: when a charged particle flies through, it generates an electrical signal. But sometimes the chamber fires a false hit, as if an invisible hand pressed a doorbell. These spurious signals muddy the picture and make it harder to pinpoint the particle's path.
The researchers collected data from one RPC chamber and three auxiliary scintillators that served as a reliable reference. For each signal, they computed 15 simple numbers: cluster size (how many neighboring points fired), width and mean of the time and charge distributions. These numbers are like a person's height, weight, and gait: they help tell a real track from a fake one. Then three different machine learning algorithms learned from these features: a deep neural network, a one-dimensional convolutional network, and gradient boosting with XGBoost.
All three models performed well, but XGBoost proved the most robust on new data. The analysis showed that cluster size and the temporal shape of the signal are the most important features. This means that even without complex hardware, one can quickly and accurately filter out noise. Such an approach is useful for detectors that operate without an external trigger and must decide on their own what counts as a real event.
Resistive plate chambers were invented in the 1980s and have since been used in major experiments such as ATLAS and CMS at the Large Hadron Collider.