Análisis comparativo de técnicas de inteligencia artificial para la detección de la violencia escolar

Abstract

Introduction: School violence represents a global issue that affects the physical, emotional, and academic well-being of students. It can manifest itself in various forms, including verbal, physical, social, digital, socioeconomic, and sexual abuse. Faced with the growing need for more effective strategies for its prevention and detection, Artificial Intelligence (AI) emerges as a promising tool to address this challenge in educational contexts.

Objective: To analyze and compare different artificial intelligence techniques applied to the early detection of school violence, evaluating their effectiveness and potential integration in hybrid systems.

Method: A systematic literature review was done following the PRISMA protocol and using the Scopus database.

Results: The approach based on natural language processing (NLP) and machine learning achieved an F1 score of 84.2% in identifying at-risk students. The second study, focusing on voiceprint and speech recognition technologies, facilitated real-time harassment detection, although it did not report quantitative metrics. The third approach, which employed computer vision with YOLOv8 and
LSTM neural networks achieved 95.67 % accuracy in identifying violent behavior.

Conclusions: Artificial intelligence techniques applied to school violence offer complementary advantages: computer vision excels in accuracy for direct detection, PLN is helpful for early prevention, and speech recognition enables immediate responses. Integrating these methods into hybrid systems, with an ethical and collaborative approach, emerges as a comprehensive and effective solution to address this problem in the educational environment.