RESUMO
It has always been the goal of many researchers to gain a thorough understanding of the patterns in the stock market and forecast the trends it will follow. The use of an advanced forecasting model can assist with accurately forecasting the future price of stocks, their fluctuations in the markets, as well as make profits in trading. With this motivation, in this study, a novel stock index trend predictor model is designed by integrating Multiple Criteria Decision-Making (MCDM) with an optimized Online Sequential Extreme Learning Machine (OSELM). Forecasting the future stock index prices and analyzing the upward or downward trends of these price forecasts are the two objectives of the proposed model. As the performance of OSELM is heavily dependent on the activation functions used in it, suitable selection of the activation function for OSELM is addressed as a MCDM problem. According to this approach, the trend prediction performance of six popular activation functions is assessed based on five regression-based and five classification-based criteria. In this investigation, three MCDM approaches are used to assess the performance matrix and determine which activation function is the best for OSELM based on six alternative models and ten criteria. To further optimize OSELM's performance, a hybrid crow search algorithm (hCSA) is incorporated in its training phase. By introducing the chaotic map and mutation operator in position update scheme and catfish behavior in the search process of original CSA, the proposed hCSA is able to achieve the right balance between exploration and exploitation improving the convergence. The proposed trend predictor model is empirically evaluated over historical data of three stock indices such as BSE SENSEX, S&P 500 and DJIA collected during pre-COVID and COVID time frame. In most of the test cases, the hCSA-OSELM model outperforms the state-of-the-art baseline models in terms of all evaluation criteria. When compared to the second-best baseline model, the suggested model is able to achieve the MSE improvements of 4-6%, 25-31%, and accuracy improvements of 0.4-0.8%, 0.9-1.3% over the pre-COVID and COVID time-frames, respectively. The statistical test also reveals the better performance of the proposed model. The robust and reliable MCDM-based model selection, superior prediction and classification outcomes clearly reveal that the proposed model can be used for financial time-series forecasting amid daily volatility as well as highly volatile markets.
RESUMO
In many fields such as data mining, machine learning, pattern recognition and signal processing, data sets containing huge number of features are often involved. Feature selection is an essential data preprocessing technique for such high-dimensional data classification tasks. Traditional dimensionality reduction approach falls into two categories: Feature Extraction (FE) and Feature Selection (FS). Principal component analysis is an unsupervised linear FE method for projecting high-dimensional data into a low-dimensional space with minimum loss of information. It discovers the directions of maximal variances in the data. The Rough set approach to feature selection is used to discover the data dependencies and reduction in the number of attributes contained in a data set using the data alone, requiring no additional information. For selecting discriminative features from principal components, the Rough set theory can be applied jointly with PCA, which guarantees that the selected principal components will be the most adequate for classification. We call this method Rough PCA. The proposed method is successfully applied for choosing the principal features and then applying the Upper and Lower Approximations to find the reduced set of features from a gene expression data.
