LIVE EVENT DETECTION FOR PEOPLE’S SAFETY USING NLP AND DEEP LEARNING

Abstract

Ensuring public safety during large-scale live events has become a significant challenge due to increasing crowd density and potential security risks. This research proposes an intelligent Live Event Detection System for people’s safety using audio analytics and machine learning techniques. The system analyzes real-time audio streams captured from event environments to detect abnormal sounds such as explosions, sudden loud disturbances, or unusual crowd behavior that may indicate safety threats. Advanced audio signal processing techniques, including Mel-Frequency Cepstral Coefficients (MFCCs), spectral contrast, and chroma features, are used to extract meaningful acoustic features from the audio signals. These features are then processed using the Light Gradient Boosting Machine (LightGBM) classifier, which provides high computational efficiency and strong predictive performance for realtime event classification. The proposed model is evaluated using diverse datasets containing audio samples from various live event scenarios such as concerts, sports events, and emergency situations to distinguish between normal and abnormal sound patterns. Experimental results demonstrate that the LightGBM model achieves high detection accuracy, low latency, and strong scalability, making it suitable for real-time deployment in public safety systems. Furthermore, the system incorporates a feedback mechanism for continuous model improvement using newly collected audio data. The proposed approach enhances situational awareness and enables authorities to receive early warnings about potential risks, thereby facilitating timely intervention and improving overall public safety during live events.