Effectiveness of machine learning methods in detecting grooming: a systematic meta-analytic review.

Journal: Scientific reports
PMID:

Abstract

This study presents a systematic review (SR) and meta-analysis (MA) on the use of machine learning (ML) methods for detecting online grooming, a form of manipulation and child sexual abuse. The SR identified 33 studies from IEEE, Web of Science, Scopus, Springer, PubMed, and Google Scholar databases, and 11 ML methods were meta-analyzed for accuracy (ACC), precision (P), recall (R), and [Formula: see text] Score (F1). Multilayer Perceptron (MLP) demonstrated the highest accuracy (ACC=92%, p<0.001) and precision (P=81%, p<0.001), excelling in capturing complex, nonlinear patterns essential for analyzing nuanced online interactions. Support Vector Machine (SVM), with an ACC of 88% (p<0.001), achieved a balanced performance, characterized by high precision (P=86%, p<0.001), recall (R=74%, p<0.001), and the highest F1 score (0.79). SVM emerges as an effective algorithm, providing a robust balance across all metrics, emphasizing its adaptability and reliability in detecting nuanced grooming behaviors. This study is pioneering in meta-analyzing ML methods applied to the effectiveness in detecting grooming. The results highlight the efficacy of certain algorithms and contribute to the identification of online predators. A crucial aspect of cybersecurity for preventing child sexual abuse.

Authors

  • Marcelo Leiva-Bianchi
    Laboratory of Methodology, Behavior and Neuroscience, Faculty of Psychology, Talca, Chile.
  • Nicolas Castillo
    Department of Computer Science, Faculty of Engineering, Talca, Chile.
  • César A Astudillo
    Department of Computer Science, Faculty of Engineering, Talca, Chile. castudillo@utalca.cl.
  • Francisco Ahumada-Méndez
    Laboratory of Methodology, Behavior and Neuroscience, Faculty of Psychology, Talca, Chile.