Novel Model for Classifying the Toxicity of Metal Oxide Nanoparticles

Metal oxide nanoparticles (MeOxNP) are receiving increasing attention in the last few years due to their various applications in electronics, medicine, and environmental remediation. However, their potential toxicity poses significant hurdles for safe usage. Therefore, this paper aims at developing a new artificial intelligence (AI)-based model for the efficient classification of the toxicity of MeOxNP using a Dynamic Pelican Optimizer finetuned Random Forest (DPO-RF) technique. A database has been prepared considering different types of nanoparticles (NPs) such as Al2O3, CuO, Fe2O3, TiO2, and ZnO, and the most important key physicochemical attributes. This model is followed by pre-processing using handling of missing values with imputation and performing standardization by applying the Z-score normalization. Features were extracted with principal component analysis (PCA) reducing dimension while keeping the vital information associated with toxicity in this model. The applied DPO-RF based model enhanced the feature selection of this model while achieving enhanced accuracy through adaptive exploration of this model. The results reflect the valid classification of MeOxNP either as toxic or non-toxic, which implies a total accuracy of about 98.2 % for classes of toxicity, and a corresponding accuracy rate of about 98.5 % for classes of nontoxicity, which is offering some important implications for the assessment of potential risks while using the respective nanotechnology application.