A novel method for classifying glasses using deep neural networks is presented. The study uses data from the USA Forensic Science Service to classify six different varieties of glass according to their oxide level. The dataset is carefully cleaned up using preprocessing methods like Min-Max Scaling and Label Encoding to improve data quality and make it more neural network architecture compatible. The deep neural network model performs admirably in this difficult task, correctly categorizing glass samples with an accuracy of 81.40%. The analysis of the results shows that different varieties of glass behave differently, pointing out both areas of strength and improvement. In the future, the work may entail investigating more sophisticated deep learning techniques including transfer learning and attention processes, as well as broadening the dataset to include a greater range of glass samples. Moreover, validation tests conducted in the actual world and the model's implementation in industrial settings may offer important new perspectives on its usefulness. In general, this study advances automated glass categorization methods, which may have consequences for activities related to environmental sustainability, product creation, and quality assurance in a variety of sectors.