Incremental feature learning (IFL) is an approach to machine learning (ML) where a predictive model is incrementally constructed by adding descriptive features during the training process when necessary. The benefits of such an approach are improved computational efficiency and better generalization performance. This approach towards construction of a predictive model results in a dynamic optimization problem, mainly due to the increasing model architectures. The latter results in a search space whose dimensionality increases over time. This talk presents a particle swarm optimization (PSO) training algorithm for IFL in neural networks (NN). More specifically, a dynamic PSO algorithm is used to train the NN, to allow for changes in the search landscape to be detected and reacted upon. Boruto feature selection is used to rank descriptive features, with the ranks used to determine the order in which features are added to the predictive model. Results presented in this talk will show that the PSO training algorithm for IFL offer competitive performance in terms of reducing NN overfitting behavior and producing an optimal model with very good generalization abilities.