a study of hybrid competitive genetic algorithm models in stock exchange expectation

©IJCERT 61 | P a g e

ISSN (Online): 2349-7084

International Journal of Computer Engineering In Research Trends (IJCERT)

Volume 1, Issue 1, July 2014, PP 61-69

www.ijcert.org

A Study of Hybrid Competitive Genetic Algorithm

Models

In Stock Exchange Expectation

Rayapu.Swathi1, M.Tech Research Scholar,

N.Parashuram2, Assistant Professor,

Dr.S.Prem Kumar3, Head of the Department

Department of CSE, G.Pullaiah College of Engineering and Technology

JNTU Anatapur, Andhra Pradesh, India

Abstract: Stock exchange expectation (SEE) assumes an imperative part in the advanced time for

any economy which is on the improvement stage. Genetic Algorithm (GA) is an evolutionary

algorithm that is helpful for taking care of issues which are excessively difficult in nature. This

paper reviews distinctive GA shows that have been tested in securities exchange expectation with

exceptional upgrade procedures utilized with them to enhance the forecast precision. The order

is made regarding GA with two layer, multilayer, neural system variations and altered

evolutionary calculations. Through the reviewed paper it is demonstrated that the execution of

GA exceeds expectations when incorporated with other machine learning algorithms.

Keywords-classification, enhancement technique, genetic algorithm, prediction accuracy, Stock

exchange expectation

1. INTRODUCTION

Stock markets are one among the key players in determinant the economy that is that the backbone of any nation and

for that matter evens the world economy. There’s continually some risk concerned in investment within the

exchange owing to its extremely unpredictable behavior. Since exchange is actually non-parametric, dynamic, time-

variant and chaotic in nature, exchange prediction could be a difficult task [1]. Additionally to the present exchange

is littered with several macro economical factors like political events, general economic conditions, firm’s policies,

investors’ expectations, institutional investors’ selections, movement of alternative stock markets and also the

science of investors etc [2]

GA plays a vital role in predicting the exchange costs accurately. Varied researches on the applying of GA in

prediction issues have verified its benefits over alternative applied mathematics and constant quantity techniques.

http://www.ijcert.org/

A Study of Hybrid Competitive Genetic Algorithm Models



GA has the benefits like simple to grasp, simple to implement, extremely configurable, doesn't suffer from the curse

of spatial property, flexible, applicable to any downside wherever a fitness perform is outlined, simply

parallelizable, output are often was AN actual program and might be run to hurry in real time, and might be wont to

optimize additional formal models. GA additionally inherits bound limitations like stagnation of population, fast

predominance of bound people over the remainder of the population, over-fitting, give the foremost work resolution

that was evolved not essentially the best resolution, laborious to see best parameters. the restrictions of ancient GA

are often engulfed by hybrid superimposed GA in predicting the exchange behavior that inculcates the options like 2

superimposed GA, multilayered GA, neural network variants with GA; modified GA. These further benefits of GA

give prediction of exchange costs at a speedier rate with high prediction accuracy compared to ancient GA.The

purpose of this work is to review and classify the hybrid GA models to exchange prediction. The results area unit

conferred in four completely different classes. The primary class lists 2 layer hybrid GA model, description and their

model comparisons. The second lists 3 layer hybrid GA model for exchange prediction. The third presents stock

markets predictions by GA with neural network variants. The last one summarizes the changed versions of easy

genetic algorithms for exchange predictions. This work focuses on the applying of accessible GA to predict

exchange indexes.GA could also be applied to various markets to forecast the exchange indexes.

II. COMPARISON OF HYBRID GA

Improving SEE by two layer hybrid GA

In [3], a hybrid neurogenetic system for stock prognostication was planned. This hybrid system used RNN trained

by a back propagation primarily based formula, to predict obtain and hold strategy of thirty six corporations in stock

market & National Association of Securities Dealers Automated Quotations for thirteen years from 1992 to

2004. Since back propagation formula is susceptible to stand still in native minima and extremely depends on the

initial weights, the GA was accustomed optimize the NN’s weight underneath a 2-D cryptography and crossover

mechanisms. The GA is parallelized on a UNIX operating system cluster system victimization message passing

interface to scale back the time in process the mass information. The experimental results showed that the hybrid

model predicts higher (34 corporations among thirty six companies) than each GA & SVM alone in obtain and

hold strategy. In [4], a hybrid GA-SVM system was developed for predicting the longer term directions of the stock

value. a group of technical indicators that exhibits high correlation were used as input options. The GA was

accustomed choose the set of most informative input options among all the technical indicators. The chosen options

were used as inputs to SVM.

SVM minimize associate degree higher bond of generalization error. The predictions performance was measured

supported the hit magnitude relation. The results were compared with the stand alone SVM. The GA-SVM hybrid

model considerably outperformed the SVM.A hybrid machine learning system supported GA-TSA model for stock

exchange prognostication was developed [5]. The success of a commerce rate depends on selecting the most

effective parameter combination .This can be done by victimization GA .The GA sets the sub domain of the

parameters and finds close to optimum worth within the sub domain with the statistic analysis during a} very cheap

Rayapu.Swathi,N.Parashuram, Dr.S. Prem Kumar

International Journal of Computer Engineering In Research Trends (IJCERT ) 63 | P a g e

time. This hybrid model outperformed the time model alone in terms of accuracy. The accuracy is still improved by

considering varied political & economical factors that affects the stock exchange and incorporating market

specific domain information in to the system. The stock exchange indices were foretold victimization hybrid

GAPSO with perturbation term galvanized by the passive congregation biological mechanism that allows all

particles within the swarm to perform the worldwide search within the whole search house [6]. The GA is employed

to extend the range and provides particles to fly within the new regions in search house. This hybrid model will

increase the prediction accuracy for each short term & long run stock exchange indices compared to GA/PSO

model alone. This model is utilized in alternative applications like pattern classification recognition and optimization

downside. A completely unique technique supported hybrid combination of GAARMA was applied for statistic

prognostication [7]. This hybrid model edges from the capabilities of ARMA to spot linear trends yet as GA’s ability

to get models that capture on non linear patterns from information. The GA evolves models adopting the info with

none restrictions with reference to the shape of models or coefficients. The empirical result with real stock

information confirmed that the hybrid approach to be a good rival over pure ARMA & GA. The accuracy is

still improved by hair care modification purpose detection technique & modeling techniques. In [8], a hybrid

model supported GA-RST foretold the stock value. During this novel technique multi-technical indicators were

accustomed predict stock value trends. This technique has utilized a RST formula to extract linguistic rules from the

linguistic technical indicator dataset and utilize GA to outline the extracted rules to urge higher prognostication

accuracy and stock come. The effectiveness of the planned model was verified with 2 styles of performance

evaluations, accuracy and stock come for 6 year amount of the Taiwan securities market capitalization weighted

index because the experiment dataset.

B.Improving SEE by three layer hybrid GA

An intelligent call network that measures all the qualitative events additionally to quantitative factors which will

influence the securities market were developed [9]. This novel methodology consists of three elements particularly

factors identification, qualitative model and call integration. The fuzzy city methodology was used to capture the

stock expert’s information and remodel it to the suitable format of GFNN. The ANN that considers solely

quantitative factors is outperformed by the projected system within the learning accuracy, buy-sell clarity and buy-

sell performance. Real-number committal to writing approach may be applied additionally to binary committal to

writing approach. A fusion model by combining HMM, ANN and GA to forecast money behavior was developed

[10]. These will be used for in-depth analysis of the securities market. Victimization the ANN the daily stock costs

were reworked to freelance sets of value. The GA is employed to search out the optimum initial parameters for the

HMM given the reworked observation sequences. This fusion model found variety of other information things from

the historical information, that exhibit similar securities market trends. The trained HMM is employed to spot and

find similar patterns within the historical information. The performance of the fusion model is best than the essential

model wherever solely one HMM is employed. It conjointly outperforms the popular applied math foretelling tool.

The hybrid VAR-NN-GA framework has machine-driven the choice method of prediction [11]. Power unit explore

for the correlate stocks and indicators mechanically. Input file like commerce volume, economic process rate and




also the currency rate tested with power unit analysis. The NN forecast from the relevant inputs was set by the

power unit analysis. The GA is employed to regulate the weights of every NN model. The VAR-NNGA hybrid

system outperformed the complete neural network in prediction accuracy. A novel approach GA-CPSO-SVR was

used to predict the money returns of Shanghai composite index [12]. GA is employed for feature choice and CPSO

is employed for optimizing the parameter of Foreign Intelligence Service model. CPSO methodology combines PSO

with accommodative inertia weight issue and chaotic linear search. The accuracy of the prediction may be improved

by considering different advanced looking techniques to see appropriate parameters and also the range of input file.

This hybrid model outperforms BPNN, ARIMA, Foreign Intelligence Service and CPSO-SVR in foretelling money

returns.

Fig 1. Genetic Algorithm layers

C.Improving SEE by GA with neural network variants

TDNN & GA were used along to urge additional correct forecast of European choice costs of groovy index

[13]. Since back propagation algorithmic rule is at risk of bog down in native minima and specifying the design at

the beginning of the algorithmic rule might not be optimum, the GA is employed to optimize the TDNN. This

approach has mounted design and coaching of TDNN was through with GA. The experimental results showed that

the hybrid model predicts higher than easy feed forward neural network. A GMDHNN and genetic algorithmic rule

was developed for stock prediction of cement sector in Iranian capital stock market [14]. Mistreatment GMDHNN a

model might be depicted as a collection of neurons within which completely different pairs of them in every layer

were connected through a multinomial and thus manufacture new neurons within the next layer. In GMDHNN the

identification method used the appropriate optimization technique GA to seek out the simplest specification. The

complete design of the GMDHNN is intended with the assistance of GA. It provides the optimum variety of

neuronal in every hidden layer and their connectivity configuration to seek out the optimum set of appropriate

coefficients of quadratic expressions to model stock costs.GMDHNN outperforms ancient statistic technique and

regression based mostly models in prediction accuracy. In [15], the author has investigated the effectiveness of a

hybrid approach supported the ATNN and therefore the TDNN with the GA in a very temporal pattern for exchange

prediction tasks.TDNN permits solely weights to be adopted however time delays are enclosed however are



mounted. The ATNN network adapts its time delay and weights throughout coaching to rise accommodate dynamic

temporal patterns and supply additional flexibility for optimization tasks.GA supports optimization of the quantity of

your time delay and network study factors at the same time for the ATNN and TDNN model. The result showed that

the accuracy of this approach is over that of the quality ATNN, TDNN and RNN.

D.Improving SEE by modified GA

TIMTAEF method, which performs an evolutionary search for the minimum dimension in determining the

characteristic phase space that generates the financial time series was presented [16].This hybrid model composed of

modular morphological neural network with a modified genetic algorithm(MGA).The modified genetic algorithm

was used to improve search convergence. The prediction of the proposed model obtained a performance much better

in terms of evaluation function than the TAEF & MRLTAEF model To increase prediction accuracy and reduce

search space and time for achieving the optimal solution, the combination of WNN with fuzzy knowledge was used

[17]. The proposed FWNN structure is trained with differential evaluation (DE) algorithm. DE includes evolution

strategies and conventional GA. DE is used for minimizing non-linear and no differentiable continuous space

function. Training FWNN system by DE is much more faster than the traditional GA.The result, demonstrated that

FWNN with DE has better performance than FWNN with BP, FFNN and ANFIS.In [18], the author has investigated

the chaotic behavior of NASDAQ and S&P CNX NIFTY stock markets using EDA based LLWNN.The LLWNN

system was optimized using estimation of distribution algorithm which is a new class of evolutionary

algorithms.EDA explicitly extract global statistical information from the selected solutions. The major issues in

EDAs are selecting parents and building probability distribution model. The author used truncation selection and

Gaussian distribution with diagonal covariance matrix. The result showed that the accuracy of this approach is

higher than that of the standard WNN. Table I lists the main features of hybrid GA.An evolutionary neural network

model was proposed to improve the performance of ANN in time series forecasting [19]. A modified GA was used

to select best input features, optimize the slope of hidden modes activation function, learning parameters and also

the number of hidden layer nodes. To evaluate the effectiveness of the proposed model FOREX rate prediction acts

as a benchmark application. The result showed that the accuracy of this approach is higher than that of hybrid RNN-

ARIMA, fuzzy system-ANN and ANN.

TABLE I. MAIN FEATURES OF HYBRID GA

Hybrid GA Main features

GA-FS

reduce the processing time without loss of performance, , reduce computational cost using message

passing interface, improves the

generalizations, avoids the curse of dimensionality, avoids local minima

GA-SVM resistance to over fitting problem, optimal selection of most informative input features from the

technical indicator




GA-TSA finds best parameter combination which is the key for successful trading

GA-PSO

fast convergence ,robustness

GA-ARMA more robust, accurate model of time series

GA-RST

extract rules effectively from past stock data, improve classification accuracy and forecasts profit

,produces more reasonable &Understandable rules, provide objective suggestions

GA-FS-NN X GA-FS-NN Captures the stock expert’s knowledge, decreases the training time and avoids local

minima

ATNN

&TDNNGA

supports the optimization of the number of time delays & network architectural factors

simultaneously, adapts both time-delay and weights to accommodate changing temporal patterns

and provides more flexibility

MGA based

MMNN

Overcomes the random walk dilemma for stock market prediction, does not discard any possible

correlation that exist among the time series parameters, even higher order correlations does not

make any prior assumption

DE based

FWNN

Combines the strengths of wavelet theory, fuzzy logic and neural networks, fast training , ability to

analyze non-stationary signals to discover their local details, self learning characteristic that

increases the accuracy of the prediction, less time for parameter updating

EDA based

LLWNN

Learning efficiency, structure efficiency, provides more parsimonious interpolation in high

dimension spaces when modeling samples are sparse

MGA based

evolutionary

NN

Selects best input features,optimizes the slope of the hidden nodes activation function,learning

parameters and the number of hidden layer nodes.

III. CONCLUSION

This study has surveyed articles that have applied hybrid GA model to predict stock exchange values. This study has

focused on the outline of hybrid GA, model comparison and its increased options. Ancient GA has its own

drawbacks thereby limiting its performance to a precise level of accuracy. The limitation of ancient GA is overcome

by hybrid stratified GA in predicting the stock exchange behavior that inculcates options like 2 layer, multilayer

& changed organic process algorithms.

APPENDIX : ABBREVIATION

ANFIS : Adaptive neuro fuzzy inference system

ANN Artificial neural network

ARMA Auto regressive moving average

ATNN Adaptive time-delay neural network

CPSO Chaotic particle swarm optimization



DE Differential evolution

EDA Estimation of distribution algorithms

FFNN Feed forward neural network

FWNN Fuzzy wavelet neural network

GFNN Genetic fuzzy neural network

GMDHNN Group method of data handlingneural network

HMM Hidden markov model

LLWNN Local linear wavelet neural network

MGA Modified genetic algorithm

MRLTAEF morphological rank linear time delay added evolutionary forecasting

PSO Particle swarm optimizitation

RNN Recurrent neural network

RST Rough set theory

SVM Support vector machine

SVR Support vector regression

TAEF Time-delay added evolutionary forecasting

TDNN Time delay neural network

TIMTAEF Translation invariant morphological time-added evolutionary forecasting

TSA Time series analysis

VAR Vector auto regression

WNN Wavelet neural network

REFERENCES

[1] M. T. Leung, H. Daouk and A. S. Chen, “forecasting stock indices: A comparison of classification and level

estimation models,” International Journal of Forecasting 16, pp. 173-190, 2000.

[2] K. Manish and M. Thenmozhi, “Forecasting stock index movement: A comparision of support vector machines

and random forest,”Proceedings of Ninth Indian Institute of Capital Market Conference,India, Jan 2006.

[3] Y. Kwon and B. Moon, “A hybrid neurogenetic approach for stock forecasting,” IEEE Transactions on Neural

Networks, vol. 18, no. 3, pp.851-864, May 2007.

[4] R.Choudhry and K.Garg, “A hybrid neurogenetic approach for stock forecasting,” World Academy of

Science,Engineering and Technology 15, pp. 315-318, 2008.

[5] L. Kumar, A. Pandy, S. Srivastava and M. Darhari, “A hybrid machine learning system for stock market

forecasting,” Journal of International Technology and Information Management, vol. 20, no. 1, pp. 39-47,Jan 2011.

[6] T. Aboueldahab and M. Fakhreldin, “Prediction of stock market indices using hybrid genetic algorithm/particle

swarm optimization with perturbation term,” International Conference on Swarm Intelligence,June 2011.

[7] A. Barbulescu and E. Bautu, “A hybrid approach for modeling financial time series,” The International Arab

Journal of Information Technology, vol. 9, no. 4, pp. 327-335, July 2012.




[8] C. H. Cheng, T. L. Chen and L. Y. Wei, “A hybrid model based on rough sets theory and genetic algorithms for

stock price forecasting,”Information Sciences 180, pp. 1610-1629, 2010.

[9] R. J. Kuo, C. H. Chen and Y. C. Hwang, “An intelligent stock trading decision support system through

integration of genetic algorithm based fuzzy neural network and artificial neural network,” Fuzzy Sets and Systems,

vol. 118, pp. 2145, 2001.

[10] M. R. Hassan, B. Nath and M. Kirley, “A fusion model of HMM, ANN and GA for stock market forecasting,”

Expert Sytems with Applications,vol. 33, pp. 171-180, 2007.

[11] S. I. Ao, “A framework for neural network to make business forecasting with hybrid VAR and GA

components,” Engineering Letters, 13:1, pp.24-29, May 2006.

[12] Y. Fu and Y. Cheng, “Application of an integrated support vector regression method in prediction of financial

returns,” International Journal of Information Engineering and Electronic Business, vol. 3, no.3, pp. 37-43, June

2011.

[13] P. K. Mishra, P. Kumar, V. Sharma and J. Dhar, “Option pricing using hybrid neuro evolution approach,”

International Research Journal of Finance and Economics, issue 64, pp.115-125, April 2011.

[14] S. Fallahi, M. Shaverdi and V. Bashiri, “Applying GMDH-type neural network and genetic algorithm for stock

price prediction of Iranian cement factor,” Applications and Applied Mathematics: An International Journal, vol. 6,

no. 2, pp. 572-591, Dec 2011.

[15] H. Kim and K. Shin, “A hybrid approach based on neural networks and genetic algorithms for detecting

temporal patterns in stock markets,”Applied Soft Computing 7, pp. 569-576, 2007.

[16] R A. Araujo, “Translation invariant morphological time-lag added evolutionary forecasting method for stock

market prediction,” Expert Systems with Applications 38, pp. 28352848, 2011.

[17] R. H. Abiyev and V. H. Abiyev, “Differential evaluation learning of fuzzy wavelet neural networks for stock

price predictions,” Journal of Information and Computing Science, vol. 7, no. 2, pp.121-130, 2012.

[18] Y. Chen, X. Dong and Y. Zhao, “Stock index modeling using EDA dased local linear wavelet neural network,”

IEEE, pp.1646-1650, 2005.[19] M. Sheikhan, B. Movaghar, “Improvement of time series forecasting using modified

GA neural hybrid architecture,” Majlesi Journal of Multimedia