Comparative Analysis of Predictive Models for Regeneration Capability of a Liquid Desiccant Regenerator: Regression Tree vs. Artificial Neural Network

In the current study, a predictive model using an artificial neural network and a Regression Tree is modeled to predict the regeneration rate of a liquid desiccant regenerator corresponding to a particular size and operating conditions. The effect of the parameters of the individual models such as the number of layers, the number of neurons, and the level of pruning on the prediction performance and the training time is investigated. It has been observed that the average accuracy of the artificial neural network is about 98.19% which is higher than the accuracy of the Regression Tree (85.23%) but the training time of the Regression Tree is 0.0051 s which is much shorter than the time required to train the artificial neural network (0.6995 s). It is also observed that the Regression Tree is found to be more robust and reliable when the input to the model is not in the close vicinity of the training data. On the contrary, the artificial neural network has higher accuracy than the Regression Tree when the input to the model is in close vicinity of the training data. The developed models aim to reduce the time required for the initial design of the liquid desiccant regenerator by providing a faster response than traditional numerical models.