A novel algorithm for model evaluation in acoustic studies with small datasets: A comparison of linear and neural network models

Traditional neural network models rely on dividing data into training and testing sets to evaluate performance. However, when working with small datasets—a common challenge in acoustic studies—the test model's sensitivity to the selection of training and testing data poses a significant limitation, often leading to inconsistent results. To address this issue, we propose a novel algorithm that provides an alternative structure for model evaluation. This study introduces the proposed algorithm and demonstrates its effectiveness by applying it to a small acoustic dataset. As a case study, we use the algorithm to compare the efficacy of a linear model with an artificial neural network model. Our results highlight the potential of this approach to improve reliability in neural network applications within the field of acoustics, especially when data availability is limited.