High-performance glass classification using advanced machine learning and deep learning algorithms with a comprehensive feature analysis

Glass classification with accuracy is highly required in construction, automotive, and electronics industries, where material properties like transparency and strength are vital. Traditional practices, though effective, are time-consuming and non-scalable. This paper proposes a solution based on Machine Learning and Deep Learning to automate and scale up the accuracy of glass classification. The work uses a dataset of 214 samples with nine chemical and physical properties. Exploratory Data Analysis provides significant patterns and verifies pre-determined glass classes through clustering techniques like Gaussian Mixture Models. Advanced learning algorithms like Random Forest (RF), XGBoost, Support Vector Machines, and Bidirectional Long Short-Term Memory (BiLSTM) networks are applied for classification. Findings prove RF and XGBoost to provide the highest classification accuracy, and BiLSTM to be the best in recognizing complex data patterns. Feature importance analysis pinpoints significant features and identifies magnesium and barium among those used to distinguish between glass types. This detailed evaluation highlights the potential of AI-based methods to revolutionize classifying glass with increased accuracy, efficacy, and material details.