Algorithm |
Method |
Advantage |
Challenge |
Extreme Learning Machine (ELM) [28] |
Reviews the theoretical model of different ELM algorithms |
High stability, speed, and accuracy under general or specific conditions
Supports many learning applications, such as regression, classification, feature selection, clustering, and representational learning |
Improves a data-dependent generalization mechanism for generating hidden-layer parameters |
Artificial Neural Networks (ANNs) [29] |
Demonstrate LM–ANN models to analyze and predict the output given by a data set |
Compare different learning strategies |
Generalization and optimization capabilities of the learning system |
Online machine learning [30] |
Online NN, Online Support Vector Machine (SVM), online Kernel Principal Component Analysis (KPCA) |
The classifier can adapt or retrain the changes in the input data for prediction.
The prediction and online classification processes are sometimes integrated for big data analytics. |
An extensive demonstration in practical applications remains a significant challenge for online-learning methods. |
ELM Clustering (ELMC) [31] |
KMeans algorithm in ELM, non-negative matrix factorization (NMF) algorithm in ELM |
An ELM feature space is supported by KMeans clustering.
It can handle a large number of input parameters.
NMF is tested for finding the low-dimensional representation of non-negative high-dimensional data, which can provide initiative data mapping to simplify the process.
The overall performance has less effect with the changes in the hidden-layer nodes. |
Further testing needs to be conducted, especially with other ELM feature-mapping techniques. |
Unsupervised Discriminative Extreme Learning Machine (UDELM) [32] |
Handles learning tasks with only unlabeled data |
Merges the local manifold learning with global discriminative learning
Gives better data representation than the ordinary unsupervised ELM, which conserves only the local structure of data |
Generalization and optimization factors need to be enhanced. |