Classification of Glass Artifacts Using the Black Widow Optimized Backpropagation Neural Network Algorithm

By optimizing the selection of initial weights and bias parameters, as well as the Adam optimizer and network architectural parameters using BWOA, we addressed issues related to local optima and overfitting encountered in the BP optimization process. Subsequently, within the context of classifying ancient glass artifacts, we compared the performance of the BWOA-optimized BP neural network with that of the Genetic Algorithm (GA) BP neural network and the traditional BP neural network. Our results demonstrated that the BWOA-optimized BP neural network achieved an average prediction error of 14.6% on the test set. Notably, the BWOA-optimized BP achieved ideal adaptability in the second iteration, 83 generations earlier than the GA-BP. The average absolute error (MAE), root mean square error (RMSE), mean square error (MSE), and mean absolute percentage error (MAPE) of the BWO-BP on the test set were calculated to be 1.442, 1.977, 3.909, and 0.145, respectively, outperforming the other two algorithms. Overall, compared to the traditional BP neural network, the BWOA-BP neural network significantly improved prediction accuracy and had a higher convergence rate, thereby providing valuable insights for the optimization and enhancement of BP neural networks.