Optimizing Real-time Signal Processing in Particle Physics Using Machine Learning Algorithms for High-energy Particle Detection

Abstract

Real-time signal processing is essential in high-energy particle physics, where detector systems produce large volumes of waveform data under strict latency and storage constraints. This study examines the use of machine learning algorithms for waveform-based event classification in high-energy particle detection, with emphasis on signal-background discrimination and trigger-oriented processing. Detector responses are represented as one-dimensional time-series waveforms that encode temporal structure, amplitude variation, noise components and non-linear behaviour. The manuscript evaluates the relevance of convolutional neural networks, long short-term memory networks and a hybrid CNN-LSTM approach for extracting discriminative features from these complex signals. The analysed dataset contains labelled signal and background events, allowing supervised classification after preprocessing steps such as normalisation, filtering, outlier treatment and interpolation to improve signal quality and model robustness. The proposed CNN-LSTM model correctly identified 116 of 120 signal events and 115 of 120 background events in the test set, with a reported accuracy of 96.25%. Comparative performance indicates that the hybrid model outperformed decision tree, random forest, support vector machine, CNN and LSTM models in the reported analysis. These findings suggest that deep learning can support efficient waveform classification and may assist real-time trigger decisions when integrated with suitable low-latency hardware frameworks. However, further validation using larger benchmark datasets, transparent training protocols, hardware-based latency assessments across different event classes, noise levels, and realistic detector operating conditions is required before operational implementation in particle-physics experiments.