A TOPSIS-ELM framework for stock index price movement prediction

Abstract

In recent years Extreme Learning Machine (ELM) has gained the interest of various researchers due to its superior generalization and approximation capability. The network architecture and type of activation functions are the two important factors that influence the performance of an ELM. Hence in this study, a Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) oriented multi-criteria decision making (MCDM) framework is suggested for analyzing various ELM models developed with distinct activation functions with respect to sixteen evaluation criteria. Evaluating the performance of the ELM with respect to multiple criteria instead of single criterion can help in designing a more robust network. The proposed framework is used as a binary classification system for pursuing the problem of stock index price movement prediction. The model is empirically evaluated by using historical data of three stock indices such as BSE SENSEX, S&P 500 and NIFTY 50. The empirical study has disclosed promising results by evaluating ELM with different activation functions as well as multiple criteria.

Developing a novel stock index trend predictor model by integrating multiple criteria decision-making with an optimized online sequential extreme learning machine

“…A systematic survey on recent trends in stock market prediction by Jiang ( 2021 ) revealed that researchers have used various classification as well as regression metrics to evaluate the performance of stock market prediction models. As suggested by (Kou et al 2012 ; Samal and Dash 2021a ), if the performance of a model is evaluated over multiple criteria, a robust performance evaluation system can be generated. To empirically authenticate their proposal, the authors (Kou et al 2012 ) evaluated the performance of 11 different classification schemes over 10 performance evaluation criteria using MCDM techniques such as TOPSIS, ELECTRE, grey relational analysis, VIKOR, PROMETHEE.…”

“…The weights of individual base classifiers were assigned based on their respective TOPSIS ranking. More recently in Samal and Dash ( 2021a ), 15 different activation function-based ELM frameworks have been evaluated based on 16 criteria for the prediction of stock index price movements. It is reported that MCDM-based evaluation of classifiers can provide a more robust and reliable outcome.…”

“…In the literature, researchers (Das et al 2019 ; Yu et al 2020 ) have only employed widespread activation functions such as Sigmoid, Gaussian, Tanh, and others to predict financial time series using ELM and its variations. According to a recent study (Samal and Dash 2021a ), no single activation function can achieve excellent performance for diverse datasets, and linear unit-based activation functions have a higher overall generalization capability for stock market prediction.…”

“…Although OSELM outperforms basic ELM in terms of performance, the selection of an appropriate activation function for OSELM's hidden layer has not been discussed in the literature, and only popular activation functions such as Sigmoid and Gaussian have been used for prediction of financial time-series data. Certain research (Samal and Dash 2021a ), on the other hand, imply that activation functions of the Linear unit family (ELU, ReLU, LReLU, PReLU) can produce results comparable or better than popular activation functions. As a result, the determination of activation function necessitates additional analysis.…”

“…The input layer weights in a basic ELM network were assigned arbitrarily, whereas the output layer weights were calculated mathematically. Since its inception, ELM has seen a wide range of applications, including electrical load prediction (Zhang et al 2013), financial market prediction (Das et al 2020;Dash et al 2014;Dash and Dash 2015b;Samal and Dash 2021a), sales market analysis (Sun et al 2008) and many others (Huang et al 2011). However, a preliminary survey of the literature reveals a number of significant challenges relating to ELM: (a) Single pass non-iterative training may not produce efficient results; (b) The incorporation of random, non-optimal input weights may have an influence on outputs.…”