Bijan Bihari Misra

Index prediction with neuro-genetic hybrid network: A comparative analysis of performance

Forecasting stock exchange rates is an important financial problem that is receiving increasing attention. During the last few years, a number of neural network models and hybrid models have been proposed for obtaining accurate prediction results, in an attempt to outperform the traditional linear and nonlinear approaches. This paper evaluates the effectiveness of neural network models which are known to be dynamic and effective in stock-market predictions. The models analyzed are artificial neural network (ANN) trained with gradient descent (GD) technique, ANN trained with genetic algorithm (GA) and functional link neural network (FLANN) trained with GA. The stock price index of Bombay stock exchange data has been considered to train these models and to compare their relative performance. Experimental results and analysis has been presented to show the performance of different models.

Evaluation of normalization methods on neuro-genetic models for stock index forecasting

With the rise of artificial intelligence technology and the growing interrelated markets of the last two decades offering unprecedented trading opportunities, technical analysis simply based on forecasting models is no longer enough. To meet the trading challenge in todays global market, technical analysis must be redefined. Before using the neural network models some issues such as data preprocessing, network architecture and learning parameters are to be considered. Data normalization is a fundamental data preprocessing step for learning from data before feeding to the Artificial Neural Network (ANN). Finding an appropriate method to normalize time series data is not a simple task. This work evaluates various normalization methods used in ANN model trained with gradient descent (ANN-GD), genetic algorithm (ANN-GA), and functional link artificial neural network model trained with GD (FLANN-GD) and genetic algorithm (FLANN-GA). The study is applied on daily closing price of Bombay stock exchange (BSE) and experimental result.

Hybridzing chemical reaction optimization and Artificial Neural Network for stock future index forecasting

Stock index forecasting has been a cornerstone and challenging task in computational statistics and financial mathematics since last few decades. Several machine learning methods have been proposed in order to forecast the future value of stocks effectively as well as efficiently. In this paper we considered an Artificial Neural Network (ANN) combined with a Chemical Reaction Optimization (CRO) algorithm forming a hybridized model (ANN-CRO) to forecast the Bombay Stock Exchange (BSE) future indices. Uniform population method (UP) has been used as initial population for CRO. The preprocessed data which includes the daily closing prices of BS E have been used for training and testing purpose. The predictability performance of the model is evaluated in terms of Average Percentage of Errors (APE), and compared with the result obtained by using a multilayer perceptron (MLP) model. It may be concluded that the ANN-CRO model can be a promising tool for the purpose of stock index prediction.

Pipelining the ranking techniques for microarray data classification

Layout of the pipelined rank based microarray data classification.Display Omitted This is a technique for feature selection and classification of microarray databases.Here rather than choosing single ranking method number of pipeline of ranking methods are used.Using different classifier a stable pipeline for feature selection is chosen. Identification of relevant genes from microarray data is an apparent need in many applications. For such identification different ranking techniques with different evaluation criterion are used, which usually assign different ranks to the same gene. As a result, different techniques identify different gene subsets, which may not be the set of significant genes. To overcome such problems, in this study pipelining the ranking techniques is suggested. In each stage of pipeline, few of the lower ranked features are eliminated and at the end a relatively good subset of feature is preserved. However, the order in which the ranking techniques are used in the pipeline is important to ensure that the significant genes are preserved in the final subset. For this experimental study, twenty four unique pipeline models are generated out of four gene ranking strategies. These pipelines are tested with seven different microarray databases to find the suitable pipeline for such task. Further the gene subset obtained is tested with four classifiers and four performance metrics are evaluated. No single pipeline dominates other pipelines in performance; therefore a grading system is applied to the results of these pipelines to find out a consistent model. The finding of grading system that a pipeline model is significant is also established by Nemenyi post-hoc hypothetical test. Performance of this pipeline model is compared with four ranking techniques, though its performance is not superior always but majority of time it yields better results and can be suggested as a consistent model. However it requires more computational time in comparison to single ranking techniques.

An Adaptive Second Order Neural Network with Genetic-Algorithm-based Training (ASONN-GA) to Forecast the Closing Prices of the Stock Market

Successful prediction of stock indices could yield significant profit and hence require an efficient prediction system. Higher order neural networks (HONN) have several advantages over traditional neural networks such as stronger approximation, higher fault tolerance capacity and faster convergence characteristics. This paper proposes an adaptive single layer second order neural network with genetic algorithm based training (ASONN-GA) applied to forecast daily closing prices of the stock market. For comparative study of performance, two conventional neural based models such as a recurrent neural network (RNN) and a multilayer perceptron (MLP) have been developed. The optimal network parameters for all the three models are tuned by genetic algorithm (GA). The efficiencies of the models have been evaluated by forecasting the one-day-ahead closing prices of real stock markets. From simulation studies, it is revealed that the ASONN-GA model achieve better forecasting accuracy over other two models.

Efficient Microarray Data Classification with Three-Stage Dimensionality Reduction

High dimensionality and small sample size are the intrinsic nature of microarray data, which require effective computational methods to discover useful knowledge from it. Classification of microarray data is one of the important tasks in this field of work. Representation of the search space with thousands of genes makes this work much complex and difficult to classify efficiently. In this work, three different stages have been adopted to handle the crush of dimensionality and classify the microarray data. At the first stage, statistical measures are used to remove genes that do not contribute for classification. In the second stage, more noisy genes are removed by considering signal-to-noise ratio (SNR). In the third stage, principal component analysis (PCA) method is used to further reduce the dimension. Finally, these reduced datasets are presented to different classification techniques to evaluate their performance. Here, four different classification algorithms are used such as artificial neural network (ANN), naive Bayesian classifier, multiple linear regression (MLR), and k-nearest neighbor (k-NN) to validate the benefits of three-stage dimensionality reduction. The experimental results show that the use of statistical methods, SNR, and PCA improves the overall performance of the classifiers.

Gene selection and classification of microarray data: A Pareto DE approach

Sample classification is a most critical task in microarray data analysis. But representation of microarray data with the huge search space of thousands of gene makes this work more complex and difficult. To handle this problem both an efficient gene selection technique and efficient classifier is required. In this paper, we propose a multi-criterion Pareto differential evolution technique for feature selection. This technique first uses a wrapper technique i.e. a population based differential evolution gene selection (DEGS) algorithm for feature selection. The motivation of choosing differential evolution as compared to other learning technique is it tries to assign optimal ranks to each gene using probability distribution factor present in the microarray dataset using classification error as the fitness function. It is observed that these selections contain relevant genes as well as some irrelevant genes. So in the second phase bi-objective filter technique, called as Pareto based optimization is used to select minimum number of top ranked genes in the feature selection. Here we have considered information gain (IG) and Signal to noise ratio (SNR) as two objective functions for Pareto optimization. To verify the importance and relevance of the selected genes, classification using K-nearest neighbour (KNN), naïve Bayesian classifier (NB), artificial neural network (ANN) and support vector machine (SVM) is done. Our experiment is conducted over four well known microarray dataset. The experimental work shows that the proposed method is better than the existing searching method in terms of both classification error and predicted feature sets. The classification result shows that performance of SVM classifier is better than the result obtained using KNN, NB and ANN. Finally this method highlights its performance in terms of both relevance of genes and classification output.

Reduction Combination Determination for Efficient Microarray Data Classification with Three Stage Dimensionality Reduction Approach

Classification of microarray data with high dimension and small sample size is a complex task. This work explores the optimal search space appropriate for classification. Here the crush of dimensionality is handled with a three stages dimension reduction technique. At the first stage, statistical measures are used to remove genes that do not contribute for classification. In the second stage, more noisy genes are removed by considering signal to noise ratio (SNR). In the third stage, principal component analysis (PCA) method is used to further reduce the dimension. Further, how much to reduce at each stage is crucial to develop an efficient classifier. Combination of different proportion of reduction at each stage is considered in this study to find appropriate combination for each dataset which maximizes the classifier performance. Help of naive Bayes classifier is taken here to find appropriate combination of reduction.

Efficient financial time series prediction with evolutionary virtual data position exploration

Prediction of stock index remains a challenging task of the financial time series prediction process. Random fluctuations in the stock index make it difficult to predict. Usually the time series prediction is based on the observations of past trend over a period of time. In general, the curve the time series data follows has a linear part and a non-linear part. Prediction of the linear part with past history is not a difficult task, but the prediction of non linear segments is difficult. Though different non-linear prediction models are in use, but their prediction accuracy does not improve beyond a certain level. It is observed that close enough data positions are more informative where as far away data positions mislead prediction of such non linear segments. Apart from the existing data positions, exploration of few more close enough data positions enhance the prediction accuracy of the non-linear segments significantly. In this study, an evolutionary virtual data position (EVDP) exploration method for financial time series is proposed. It uses multilayer perceptron and genetic algorithm to build this model. Performance of the proposed model is compared with three deterministic methods such as linear, Lagrange and Taylor interpolation as well as two stochastic methods such as Uniform and Gaussian method. Ten different stock indices from across the globe are used for this experiment and it is observed that in majority of the cases performance of the proposed EVDP exploration method is better. Some stylized facts exhibited by the financial time series are also documented.

Fluctuation prediction of stock market index by adaptive evolutionary higher order neural networks

The stock market is complex and dynamic in nature, and has been a subject of research for modelling its random fluctuations. Higher order neural network (HONN) has the ability to expand the input representation space, perform high learning capabilities and have been utilised to solve many complex data mining problems. To capture the extreme volatility, nonlinearity and uncertainty associated with stock data, this paper compares two adaptive evolutionary optimisation-based Pi-Sigma neural networks (AE-PSNN), for prediction of closing prices of five real stock markets. For this experimental study, BSE, DJIA, NASDAQ, FTSE and TAIEX stock indices are employed for short, medium and long term predictions. The performance of the AE-PSNN models has been compared with that of a gradient descent-based PSNN (GD-PSNN) model and found to be superior in terms of prediction accuracy and prediction of change in direction.