Effectiveness of Classifying Unsafe Children's Toys Using NLP, Deep Learning and Ensemble Learning

Ensuring the safety of children's toys is essential not only for consumer protection but also for safeguarding the well-being and healthy development of children. The research aims to classify unsafe children's toys available at the online marketplace using Natural Language Processing with six supervised classification algorithms, deep learning algorithms namely CNN, and ensemble learning based on the weighted mean. The study collected 500 unsafe and safe children's toys data from three different sources: the United States Consumer Product Safety Commission (CPSC), the Europe Commission Rapid Alert System (Safety Gates), and the Amazon online retail store. Among the supervised classification algorithms for text classification, Logistic Regression attains 87% which leads to the highest accuracy while CNN for image classification achieves 74
^%
• The findings demonstrate that ensemble learning achieves 80% accuracy which outperforms other classification techniques and proves more robust and effective for children's toy classification, yielding comprehensive and insightful results.