An Efficient Implementation of Artificial Neural Networks with K-fold Cross-validation for Process Optimization

Abstract

A novel approach for building an Artificial Neural Network (ANN) to reconstruct experimental design data using Levenberg-Marquardt optimizer and validating obtained model based on K-fold cross-validation is implemented. In this approach it matters less how the data gets divided, every data point gets to be in test set precisely once and gets to be in a training set (k-1) times. Further, this helps to exclude overfitting of the model on training data and better predictions over unseen data. Moreover, this is the most significant strength and advantage of this approach. Also, by using this approach for validation and Levenberg-Marquardt optimizer two model layers, namely, the input layer consisting of multivariable nodes and the output layer consisting output node, will be conceptually created and validated. The internal layers will be incorporated based on the complexity of the problem. The performance of the obtained model is evaluated using the coefficient of determination. Besides, it is found to have excellent correspondence with experimental results. The method is also compared with existing methods based on model validation, and it shows the much-improved capability to predict optimal results.