Residential Rental Applications Screening: A Comparative Performance of Feedforward And Recursive Neural Networks Architectures

Abstract

This paper examines the application of Neural Network architectures and training algorithms in the rental application screening within the Nigerian property market. The study compares the performance of two ANN architectures, Feedforward Neural Network (FFNN) and Recursive Neural Network (RNN), using ten selected training algorithms. The research utilizes 724 datasets obtained from 53 professional property managers operating in the Lagos Metropolitan rental market. The datasets are divided into training (70%), validation (15%), and testing (15%) sets. Performance measurements used include, sensitivity, specificity, precision, geometric mean, F1 score, F2 score, Adjusted F2 score, Mathews Correction Coefficient, and Area under the curve (AUC). The results indicate that both FFNN and RNN, trained by the selected algorithms, demonstrate satisfactory performance. However, the Bayesian regularization (BR) training algorithm outperforms the other nine algorithms. On the other hand, the Gradient Descent Algorithm (GD), Scaled Conjugate Gradient Backpropagation (SCG), and Resilient Backpropagation (RP) show the least performance in training FFNN and RNN. The study concludes that a BR trained FFNN and RNN are more suitable for rental application selection in the Nigerian property market. The findings of this study provide valuable insights and decision-making models for property management professionals and investors in the Nigerian rental market.

Keywords

• Application Screening
• Feed-forward
• Recursive
• Training Algorithms
• Neural Networks