Generalized Regression Neural Network Optimized by Genetic Algorithm for Solving Out-of-Sample Extension Problem in Supervised Manifold Learning

With the advent of big data, massive amounts of high-dimensional data have been accumulated in many fields. The assimilation and processing of such high-dimensional data can be particularly challenging. Manifold learning offers a means for effectively dealing with this challenge. However, the results of applying manifold learning to supervised classification have remained unsatisfactory. The out-of-sample extension problem is a critical issue that must be properly solved in this regard. Genetic algorithms (GAs) have excellent global search capabilities. This paper proposes a generalized regression neural network (GRNN) optimized by a GA for the solution of the out-of-sample extension problem. The prediction performance of a GRNN mainly depends on the appropriateness of the chosen smoothing factor. The essence of the GA optimization is the determination of the optimal smoothing factor of the GRNN, the optimized form of which is subsequently used to forecast the low-dimensional embeddings of the test samples. A GA can be used to obtain a better smoothing factor in a larger search space, resulting in enhanced prediction performance. Experiments were performed to enable a detailed analysis of the important parameters that affect the performance of the proposed algorithm. The results confirmed the effectiveness of the algorithm.