Deep learning Techniques for Identification of Illicit Drug Identification in Youth

Drug consumption poses significant challenges to public health, demanding precise and timely prediction models for efficient intervention and prevention methods. Through the use of deep learning methods combined with ocular image analysis, this work proposes a novel method for predicting drug use. The objective is to develop reliable and understandable model to recognize drug users from their eye pictures. For supervised training, a varied dataset of eye pictures was gathered, including people with various drug use histories. On the dataset, various Convolutional Neural Network designs were chosen, trained, and optimized using the proper hyperparameters. This study proposal compares various Convolutional Neural Network CNN models, including InceptionV3, DenseNet169, and EfficientNetB2, in order to ascertain which one of these models gets the maximum accuracy. These models provide the average accuracy of 93.33%, 97.47% and 85.89% respectively. DenseNet169 outperforms InceptionV3 and EfficientNetB2 with highest accuracy of 97.47%. It demonstrates the highest overall performance among the three models for the given task