Classification of Unsafe Children’s Toys Based on NLP-Generated Corpus

Playing with toys is essential for children as it is an integral part of the learning process and development. However, young children especially toddlers and babies are susceptible to dangers due to unsafe toys. This research aims to classify unsafe children’s toys based on a corpus derived by Natural Language Processing (NLP) techniques using different classification algorithms. The study collected 19087 unsafe and safe children’s toys data from the Europe Commission Rapid Alert System (Safety Gates) and Amazon online retail store. Three different scenarios in feature selection: ‘Brands’, ‘Unsafe Terms,’ and a combination of ‘Brands’ and ‘Unsafe Terms’ were used for classification. ‘Brands’ did not serve a vital role in the classification of safe and unsafe children’s toys compared to the classification with ‘Unsafe Terms’. Although combining ‘Brands’ and ‘Unsafe Terms’ or only ‘Unsafe Terms’ provided nearly similar accuracy, classification with only ‘Brands’ did not provide good accuracy. The classification accuracy of k-Nearest Neighbor is 78%, Logistic Regression is 83%, Support Vector Machine is 82%, Decision Tree is 82%, Naïve Bayes is 50%, and Random Forest is 81% respectively. The results showed that Logistic Regression revealed the best performance as it outperformed other algorithms.