Machine learning-based classification of zeolite quantities for humidity and temperature control in fireworks storage safety

Abstract Accident occur in fireworks storage facilities are mainly caused by the variations in temperature and relative humidity that affect the instability of chemical mixes and hence leading to explosions. In this study, a machine learning decision-support system was suggested to estimate the right amount of zeolite to help achieve safe storage. Data concerning temperature and relative humidity were gathered from storage insert a space 5 kg, 10 kg, and 15 kg of wet and dry zeolites better as over a 10-day period. A total of 208 records were acquired from the experiments. Eight different classification algorithms available in the WEKA environment such as Logistic Regression, Support Vector Machine (SMO), Random Forest, BayesNet, k-Nearest Neighbor, J48, Multiclass Classifier, and Multilayer Perceptron were analyzed through stratified 10-fold cross-validation. Logistic Regression showed the best performance by achieving the highest classification accuracy of 78.41% followed by SMO that showed a classification accuracy of 77.24%. Humidity and state of zeolite proved to be the significant factors in the analysis. The results show the possibility of integration of zeolite-based humidity regulation and machine learning systems. Although the model requires further validation with use large datasets and deployment in the real-world situation, the suggested model provides a useful tool for decision.