The extreme learning machine (ELM) is an emerging machine learning technique [1]. ELM is based on the neural network concept and includes both single and multi-hidden-layer neural networks and kernel learning but it is not limited to only them. In ELM, there are several parameters that create the model. Some of them are Hidden nodes, and biases. These
are randomly initialized and fixed after processing without tuning them iteratively. The hidden nodes in emerging ELM
are not mandatory to be neuron alike. The only free parameters that are required to be learned are the connections or weights amid the hidden layer and the output layer. Emerging ELM is based on Artificial Neural Network (ANNs) concepts that have been broadly applied on many applications and problems such as pattern classification, regression and time series analysis [2]. ANNs have very good adaptability because of selection of model from features obtained in the input data. This makes them better candidates for the variety of problems such as optical character recognition [3], face detection [4], gene prediction [5], credit scoring and time series forecasting [6]. There are several classifications of ANNs. One of the simplest type of ANNs are feed forward neural networks (FNNs) that have been studied and used widely since the introduction of the back propagation (BP) algorithm [7]. Learning process in FNNs belong to the category of supervised learning approach. In this process of supervised learning, input and output samples are given to the network for many cycles so that the network can learn the relationship and correlation between the input and output and then test of the unknown is performed with the trained network. Apart from this, FNNs has been used in time series forecasting (TSF) [8]. TSF has been very active topic of research due to its applications and implications. For instance, TSF is used for climate forecasting, electricity load prediction, economical characteristics and GDP prediction, stock market prediction, electricity consumption, sales and supply chain forecasting and in many others [8]. Times series with the advent of Internet usage are becoming big and heterogeneous which opens new opportunities and challenges in terms of accuracy and cost of prediction. Section 2, we review the various applications and variants in ELM at present and their progress and development through literature. Section 3 introduces the formulations of classical Extreme learning Machine (ELM) and Online Sequential Extreme Learning Machine (OSELM) and other relevant details. In Section 4, we conduct experiments and obtain results from time series data. Section 5 discusses the results and their interpretation and Section 6 concludes the paper.