CHAPTER 2

LITERATURE SURVEY

In the course of recent decades numerous imperative changes have occurred in the nature of
finance markets. The advancement of effective communication and trading aptitude has
expanded the extent of choice for investors. Anticipating stock return is a vital financial subject
that has pulled in scientists' attention for a long time. It includes a presumption that basic
information freely available in the past has some prescient connections to the future stock returns
(Enke & Thawornwong,2005). To mine such prediction from the accessible data, data mining
strategies can be utilized to take out the processed information from this data. Therefore, a few
researchers have concentrated on technical analysis and utilizing advanced math and science.
High level of scrutiny has been committed to the field of artificial intelligence and data mining
strategies (Wang & chan,2006).

In this section recent literature in the area of machine learning techniques and soft computing
techniques that are used to gauge stock market developments are presented. The primary
contribution of this study is to give specialists with a review of recent advancements in stock
index prediction.

2.1 Review on Machine Learning Techniques for Stock Price Prediction

Machine learning techniques intend to consequently learn and perceive patterns in huge
information. There are large numbers of well-known machine learning algorithms that can be
utilized to categorize an issue given a set of peculiarities. In this area some of these algorithms
that are especially utilized as a part of classifying stock market information into "up" or "down"
periods given a set of inputs originated through macro-economic information and technical
analysis has been presented.

2.1.1 Support Vector Machine

SVM (Vapnik et al,1999) is a learning system focused around statistical learning theory. SVMs
have been utilized in (Kim, 2003) and (Hu et al, 2009) for foreseeing financial time series, and in
(Gestel et al, 2001) and (Huang et al, 2005) to outline stock trading decision support systems.
(Huang et al, 2005) propose SVM based stock market prediction technique.They contrast its
performance with Linear Discriminant Analysis, Quadratic Discriminant Analysis, and Elman
Back-Propagation Neural Network. Next to that, they propose a combining model by
coordinating SVM with other classification method.

(Lin et al,2013) propose a SVM based stock market forecast system .This system chooses a
decent feature subset, assesses stock indicator and control over fitting on stock market inclination
anticipation. They tried this methodology on Taiwan stock market datasets and found that this
system performs well than the traditional stock market forecast system.

(Lucas Lai & James Liu,2014) investigates the Support Vector Machine and Least Square
Support Vector Machine models in stock forecasting. Three predominating anticipating systems-
General Autoregressive Conditional Heteroskedasticity (GARCH), Support Vector Regression
(SVR) and Least Square Support Vector Machine (LSSVM) are consolidated with the wavelet
kernel to structure three novel algorithms namely Wavelet-based GARCH (WL_GARCH),
Wavelet-based SVR (WL_SVR) and Wavelet-based Least Square Support Vector Machine
(WL_LSSVM) to resolve the non-parametric and non-linear financial time series issue. These
wavelet-based algorithms are then distinguished with utilizing Hang Sang Index, Dow Jones and
Shanghai Composite Index.

In (Shom Prasad Das & Sudarsan Padhy,2012) Back Propagation Technique (BP) and Support
Vector Machine Technique (SVM) have been integrated to predict future prices exchange in the
Indian stock market.Likewise, in (Hu et al,2009) and(Ding et al,2008) SVMs based on basic data
anticipate the stock crises and the financial position of the companies in the Chinese market.
(Zuoquan Zhang et al, 2012) used ε − SVM (ε − support vector machine) to build a stock price
prediction model.

Shen et al (2012) propose another forecast algorithm that makes use of the temporal among
global stock markets and different financial items to foresee the following day stock trend with
the assistance of SVM. They use same algorithm with distinctive regression algorithm to predict
the actual development in the markets. At last they build a basic trading model and contrast its
performance with the existing algorithm.
A significant pitfall of SVM for the direction forecast is that the input variables lie in a high-
dimensional feature space, extending from hundreds to thousands. The capacity of the variables
obliges a considerable measure of memory and computation time. Particularly, a stock market
comprises of a few many stocks, which prompts the high dimensionality of the variables.
Accordingly, it is of significant imperative to direct dimension reduction to procure a proficient
and discriminative representation before classification.

2.1.2 K-Nearest Neighbor (KNN)

The KNN is the basic and easy classification technique when there is little or no prior knowledge
about the data distribution (Devroye et al,1981) , (Devroye et al,1994), (Wagner &
Devroye,1997) , (Devroye & Wagner,1982) , (Javier et al,2009), in their literature, they utilizes
the k-Nearest Neighbors (k-NN) algorithm to foresee HTS and, mainly, to deal with histogram
data.(Subha & Thirupparkadal nambi,2012)anticipate the stock index movement of the popular
Indian Stock Market indices BSE-SENSEX and NSE-NIFTY using k-Nearest Neighbors
algorithm (k-NN).(Sadegh Bafandeh Imandoust & Mohammad Bolandraftar,2013) studied the
effectiveness of k-Nearest Neighbor classification method on prediction of economics and
(Suresh Kumar Sharma & Vinod Sharma,2012) utilized KNN for predicting foreign exchange
rate.

Khalid Alkhatib et al (2013) utilized kNN and non-linear regression methodology to foresee
Jordanian stock trade. They presumed that kNN is a powerful algorithm and forecasts were
almost like the stock price. An effective expectation model can be judged how it precisely
predicts rise or fall in returns progressively data.

A major pitfall with KNN approach is that it needs large memory for large training sets and the
estimation accuracy can quickly degrade when number of attributes grows.

2.1.3 Artificial Neural Network (ANN)

Artificial Neural Networks (ANN) has been indicated to be a productive tool for non-parametric
representation of information in a mixed bag of distinctive contexts where the outcome is a non-
linear function of the inputs. Recently a variety of researches have been carried out on neural
systems to foresee the stock exchange change. ANNs have drawn huge investments from a few
scientists in the stock value prediction in the previous decades. (Yanshan Wang & In-Chan
Choi,2013). The ANNs are resilient in model specification compared to parametric models,
which makes it often implemented in anticipating the financial derivatives and stock
prices.Because of the less number of limitations, these methods attain familiar face.Various NN
methods were used for most favorable feature selection to produce buy and sell signals.

Guresen et al (2011) reported the legitimacy of ANNs in stock market record expectation and
give a correlation between the Fama and French model and the ANN model in the perspective of
forecast of the Chinese stock market. They report that ANNs beat the linear models from
financial foreseeing in terms of prescient force.

Goutam Datta et al (2006) discuss the design of the Indian stock market data using Artificial
Neural Network and learned the proficiency of ANN in Bombay stock exchange.

Dase & Pawar (2010) overviewed the literature on applications of Artificial Neural Network for
stock market predictions. They found that stock index foreseeing with traditional time series
analysis was proven to be difficult, but an Artificial Neural network may be suitable for this task.

Amit Ganatret & Kosta (2010) had a main focus to construct neural network for stock market
forecast. They utilized R tool to actualize the neural network with closing price, interest rate,
global indices, turnover, and inflation as a neural network input.The paper also suggested
including news, currency rate, and crude price as input to the neural network.Additionally, an
attempt was made to construct and assess a neural network with different network parameters
and also with fundamental and technical data.

In (Jinyuan Shen et al,2007) Tapped Delay Neural Network (TDNN) with an adaptive learning
and pruning algorithm was suggested to foresee the non-sequential time serial stock indexes.
(Thenmozhi,2006) has implemented the neural network to predict the returns of the BSE in daily
basis. Multilayer perception network was utilized to fabricate the day by day return's model and
the network was prepared using Error Back Propagation algorithm. The author found that the
prescient force of the network model was impacted by the earlier day's return than the initial
three-day's inputs. The study demonstrated that acceptable results can be attained when applying
neural networks to anticipate the BSE Sensex.
Hanias et al (2012) use back-propagation neutral networks to anticipate the Athens stock index.
They utilized a design with one hidden layer comprising of seven hidden neurons and report a
decent forecast execution regarding the Mean Squared Error (MSE) up to nine days ahead.

Stelios Bekiros & Dimitris Georgoutsos (2006) examines the benefit of a trading system focused
around recurrent neural network that endeavors to anticipate the direction of progress of the
market, on account of the NASDAQ composite index.

De Faria et al (2009) differentiate the prediction execution of neural networks to the adaptive
exponential smoothing technique in the Brazilian stock market. The authors express that the
neural network performs better than the adaptive exponential smoothing technique and that
correct direction prediction hit rate reached 60%. A lot of analyses are performed that show that
the result are maintained a constant rate in all the markets. The authors finally infer that a decent
prediction performance can be utilized to create beneficial speculation strategies and that their
model is suitable for making a decision support framework for the Brazilian market.

TilaKaratne et al (2009) studied modified neural network algorithms to anticipate the actual
situation of the stock market which supports the speculators what to do in the crisis situation.

Liao & Wang (2010) inspect the statistical properties of movements in a few stock lists in the
Chinese stock market. The creators utilize a stochastic time viable neural network which uses
slacked list data as inputs and put more weight on recent perceptions and less weight on old
perceptions. It is contended that the model reflects the way speculators examine the market data
and that closer past data has a stronger influence on future market movements. The authors
report a good performance in terms of anticipating error for the time viable neural network
model.

Mahdi Pakdaman Naeini et al (2010)suggest a neural network based stock market anticipation
method. They utilize two sorts of neural networks a feed forward multilayer perception (MLP)
and an Elman recurrent network. They found that MLP neural network is much better in
foreseeing stock cost changes than Elman recurrent network and linear regression method.

The author (Smita Agarwal & Murarka,2013) present different neural network model like Radial
Basis Function NN, Layer Recurrent NN, Generalized Regression Neural Network, Back
Propagation NN which are used to predict the long term and short term share of various IT
sectors, Automobile sector, and Banking sectors. The results show that all the three models
produced better results in their own way and finally conclude that the LRNN performances better
than all other models.

In the paper (Philip M Tsang et al,2006) discussed the number of techniques for the stock market
prediction in trading community. From those techniques, neural network is considered as best
and the background of neural network and back propagation algorithm is reviewed. In addition to
that, an alert system using back propagation NN for stock buying and selling is modeled. Hong
Kong and Shanghai Banking Corporation (HSBC) is used to test this modeled system and it
achieves hit rate of 70%.

Adebiyi Ayodele et al (2012) proposed a hybrid approach which integrates the fundamental
analysis and technical analysis variables for the stock market prediction based on the previous
and current values. This improves the existing approach much efficient. This technique is much
useful to the investigators and the researchers for making the decision efficiently.

The authors (Debashish Das & Mohammad Shorif Uddin,2013) analyze the two well-known
techniques for stock market prediction namely neural network and data mining. Both are having
merits in their own way. Neural network is capable of extract meaning information from large
amount of data whereas the data mining is employed to get the future trends and pattern. Thus
the author combines these two techniques that make the reliable prediction.

Victor Devadoss & Antony Alphonnse Ligori (2013) utilize the ANN which is an efficient tool
for designing the stock value. It selects the stock value of the selected stocks under the Bombay
Stock Exchange to predict the closing value with highest accuracy.

The paper (Bjoern Krollner et al,2010) reviews the recent articles for the stock market movement
prediction in area of artificial intelligence and machine learning techniques. These articles are
separated into the field of forecasting timeframe, evaluation techniques utilized, input variables
and machine learning techniques used. Finally ANN is recognized as efficient machine learning
technique in future stock market prediction.
Hence the paper (Manjul Saini & Singh,2014) proposes the back propagation algorithm along
with the ANN is used to anticipate the stock market returns daily and also the weather condition
in Dehradun. Thus the main theme of this paper is to use soft computing techniques in the area of
stock market exchange.

The paper (Mayankumar B Patel & Sunil R Yalamalle,2014) intended the work of utilizing the
ANN techniques for the stock market movement prediction of the companies under National
Stock Exchange (NSE). The historic data is used to build a model and the outcome of that model
discovers the accuracy of the model.

2.1.4 Decision Tree for Stock Price Prediction

Decision trees are effective and well known instruments for classification and forecast. In data
mining, a decision tree is a predictive model which can be utilized to represent to both classifiers
and regression models. They are likewise valuable for investigating data to gain sight into the
associations of a substantial number of competitor input variable to the target variable. A
decision tree model comprises of a set of guidelines for isolating a substantial heterogeneous
populace into smaller, more homogeneous gatherings regarding a specific target variable. The
target variable is typically absolute and the decision tree model is utilized either to compute the
benefit that a given record has a place with each of the classifications, or to characterize the
record by allotting it to the probable class.

The research by Wu et al(2006) used decision tree technique to build the work of (Lin,2004)
where Lin tried to manipulate the filter guidelines that is to purchase while the stock price rises
k% above its previous local low and sell while it falls k% from its previous local high. The
proposed modification in (Lin,2004) was done by combining three decision variables based on
the fundamental analysis. Lin’s method outperformed the filter rule is demonstrated using an
empirical test with the electronic company’s stock in Taiwan.

According to (Wu et al,2006), in Lin’s work, the condition of clustering the trading data was
done by considering the past data alone and without considering the future data. The
investigation by (Wu et al,2006) is done to optimize the filter guidelines and Lin’s literature is
done with taking account of both the past and the future information in clustering the trading
points. To do the empirical test, the data from the NASDAQ and from the Taiwan stock market
is used. These test results proved that the Lin’s method performs better than both the filter rule
and the Lin’s method in the two stock markets.

The prototype of (Wang & chan,2006) connected the idea of serial topology and planned another
decision system, to be specific the two layer bias decision tree, for stock price forecast. The
strategy created by the authors contrasts from other studies in two regards; first is to decrease the
classification error, the decision model was altered into a bias decision model. Second, a two-
layer bias decision tree is utilized to enhance purchasing precision. The empirical results
demonstrated that the exhibited decision model delivered amazing purchasing precision, and it
fundamentally outperformed than random purchase.

Jie & Hui (2008) show a data mining method integrating attribute-oriented induction, decision
tree and information gain, which suits best for structuring the decision tree model and
preprocessing financial data for financial loss prediction.

Nair et al (2010) recommends a Decision Tree Rough Set Hybrid System for stock market
forecast demonstrating the outline and performance appraisal of a Hybrid Decision Tree- Rough
Set Based System for foreseeing the future pattern in the Bombay Stock Exchange
(BSESENSEX). This framework performs better than both the neural network based framework
and the Naïve Bayes based pattern forecast framework.

In (Nair et al,2010) a hybrid decision tree- rough set based system has been suggested for
foreseeing the future trend in the Bombay Stock Exchange (BSESENSEX). To extract features
from the historical SENSEX data, many technical indicators are used in the present literature.
Finally the paper does the two things: C4.5 decision tree is used to select the relevant features
and a rough set based system is used to enhance the rules from the extracted features.

Shweta Tiwari et al (2012) propose foreseeing future trends in stock market by decision tree
rough-set based hybrid system with HHMM. It shows a hybrid system based on decision tree
rough set, for foreseeing the trends in the Bombay Stock Exchange (BSESENSEX) along with
the Hierarchical Hidden Markov Model. It additionally shows future trends on the basis of price
income and profit. The data on accounting income when arrived at the midpoint of over years
help to anticipate the present estimation of future profits.

Tsai et al (2009) focuses on merging ANN and decision tree models as hybrid machine learning
techniques which prove better accuracy than the two models when performed separately. (Carol
Hargreaves and Yi Hao,2013) performed stock selection using technical and the fundamental
information. A separate framework is generated for the class prediction and many analytical
techniques are implemented for the stock selection and for increasing performance. The stock
selection is done by using the decision tree model.

Chen & Mu-Yen (2011) set few operating rules at Taiwan Stock Exchange Corporation (TSEC)
to improve the financial distress prediction model accuracy for which they initially gathered 100
companies’ details. To extract the suitable variables they used principal component analysis
along with the empirical experiment of 37 ratios which include financial and non-financial ratios.
Thus the decision tree (DT) classification methods and the LR techniques were used to
implement the financial prediction model. The experiments procured a fulfilling result, which
affirms for the probability and legitimacy of the proposed systems for the financial distress
prediction of already listed organizations.

2.1.5 Fuzzy Time Series Based Stock Market Prediction

Song & Chissom (1993) suggested the fuzzy time series for the stock market prediction that is
widely investigated and enhanced by many scholars, and in addition, this model is used to solve
many prediction problems of different journals. The fuzzy time series to anticipate the stock
markets (Lee, et al.,2008), (Kunhuang Huarng and Hui-Kuang Yu,2004), (Cheng, et al.,2006)
achieves improved results.

Kunhuang Huarng and Hui-Kuang Yu (2005) implemented Type 2 Fuzzy Time Series Model for
anticipating Taiwan stock index. In a Type 2 model, extra observations are needed to enrich the
fuzzy relationships obtained from Type 1 models and then to enhance forecasting performance.
(Melike Sah and Konstantin Degtiarev,2005) have suggested a method to attain improved
prediction accuracy using time-invariant fuzzy time series. The only pitfall is that it uses only
historical data in the numerical form. (Ching-Hsue Cheng et al,2007) proposed a new fuzzy time-
series method, which employs both the historic data and the future data into consideration.

Chen et al (2008) evolved high order fuzzy time-series and adjust the prediction with multi-
period adaptation model which is obtained from adaptive expectation model to improve
estimating precision. They tried the model to exploratory datasets at Taiwan Stock Exchange
Capitalization Weighted Stock Index and Hang Seng Index from first-order to fourth-order
focused around one, two, three and four–adaptation period in time-series and the adaptive
expectations model to control the estimating result to be better.

Hye-young Jung et al (2010) have proposed rearranged interval method to reflect the change of
past data and to enhance the anticipating precision of fuzzy time series. Observational
investigation demonstrates that this method in determining precision is better than existing
methods and it completely reflects the variance of past data.

Nai-yi Wang et al (2009) have suggested a method which is based on the automatic clustering
methods and 2-factor high order fuzzy time series to anticipate the temperature and the Taiwan
Futures Exchange (TAIFEX).

Teoh et al (2009) implements a hybrid model based on multi-order FTS to modify forecasting
results in order to increase forecasting accuracy by using rough sets theory to mine fuzzy logical
relationship from time-series and adaptive expectation model.

Shyi-Ming Chen and Jen-Da Shie (2009) predicted the stock price fluctuations of Taiwan Stock
Exchange to attain the optimized linguistic intervals with fuzzy time-series model by using
entropy-based discretization partitioning.

Chi - chen wang (2011) have compared the applications of the forecasting methods Auto
Regressive Integrated Moving Average (ARIMA) time series model and fuzzy time series model
by heuristic models on the amount of export values in Taiwan and (Chi - chen wang, 2011) has
attempted to use information of export values in Taiwan as an example to test whether the fuzzy
time series is indeed practical in its forecast of macroeconomic variables. By comparing fuzzy
time series with ARIMA time series method, better understanding of the appropriateness of
forecasting model could be obtained.
Tai-Liang Chen et al (2011) suggested a novel price-pattern detection method to have some
price-patterns (“price trend” and “price variation”) which includes in the time series variables to
forecast stock market.

Abhishek Gupta et al (2014) combined clustering and the classification methods together. The
dataset from the stock market is clustered using the K-means algorithm and the resultant values
from clustering are classified using horizontal partition decision tree. This hybrid method gives
close prediction of actual value which is more accurate and efficient.

2.2 Review on Evolutionary Algorithm for Stock Price Prediction

Numerous endeavors have been made to do financial market anticipation in a more experimental
way, where practically small or zero suppositions are set in a way how time series of perceptions
occur or the underlying methodology generating them. One mainstream method for doing this
has been to utilize evolutionary methods, where parameters of models or the models themselves
are emerged by utilizing systems like genetic methodologies in the biological world,
incorporating functions, for example, crossover and mutation. There have been a few
applications of Genetic Algorithms (GA) to the financial issues, for example, bankruptcy
prediction, portfolio optimization, fraud detection, financial forecasting, and scheduling (Iba and
Sasaki,2001). For various problems in the time series prediction Genetic Programming (GP) is
implemented. Genetic Programming (GP) is developed by (Koza,1992), which is a symbolic
optimization technique.GP is used to develop a one-day-ahead model in (Kaboudan,2000) to
anticipate stock prices.

Xia Pan et al (2005) employed a mutation only genetic algorithm (MOGA) to Microsoft, Intel
and Dell information from NASDAQ to scan for trading regulations that would boost gains.
They found that investment standards including purchasing, selling, holding and exchanging
between two stocks outperformed speculation standards involving a solitary stock. Additionally,
MOGA was a more effective tool than the customary methods, for example, random walk, buy
and hold or intense search.

Giving attention to the risk and desirable return of investment,(Hassan,2008) implements the
Multi-Objective Genetic Programming (MOGP) techniques for portfolio optimization in the
United Kingdom for Financial Times Stock Exchange 100 (FTSE-100) stocks between January
2002 and December 2005. The author discovers the solutions of MOGP to be non-sequential
models of financial indicators. The major two tasks of MOGP were to minimize risk which was
the standard deviation of return and to maximize return, which was the annualized average
return.

Matsui et al (2010) proposes a new assessment technique to overcome the over fitting problem in
the Genetic Algorithm training. The authors conclude that the new evaluation method performed
well than the conventional one while the results of these two assessments are compared.

Particle swarm optimization (PSO) is a robust optimization technique that is inspired by self-
organized nature of birds (Satyobroto Talukder, 2010), (Dian Palupi Rini et al,2011). Particle
Swarm Optimization has been applied for anticipation of different stock market exchanges
(Majhi Ritanjali, et al., 2008) and critical result are noted.

2.3 Hybrid Approach for Stock Market Prediction

In (Kwon et al,2005) a neuro-genetic stock anticipation framework is presented, which is

focused around the money related relationships among organizations. The genetic algorithm is
utilized to choose a set of useful information characteristics among them for a recurrent neural
network. In (Choudhry and Garg et al,2008), a hybrid GA-SVM framework was produced for
foreseeing the future direction of the stock cost. A set of specialized indicators which displays
high relationship were utilized as input characteristics. The GA was utilized to choose the set of
most useful input characteristics among all the specialized indicators. The chosen characteristics
were utilized as inputs to SVM.

A hybrid model using GA is suggested by (Rafiul Hassan et al,2006) which utilizes the power of
Hidden Markov Models (HMM), GA, and ANN to predict the financial market exchange. The
stock prices of every day are transformed to independent value sets which sends as the input to
the HMM by using ANN. The initial parameters of HMM is optimized using GA. This parameter
after optimization is used to recognize and locate similar trends in the past data.

The literature discusses the effectiveness of an integrated approach based on Genetic Algorithm
(GA) and Time Delay Neural Networks (TDNN) in (Hyun-jung Kim and Kyung-shik Shin,
2006). The GA is used to reduce the number of time delays whereas the TDNN is to obtain the
optimum prediction performance.

In (Cheng et al,2010) a combined model of GA and RST for the prediction of the stock market
price in which the multi-technical indicators were used to predict stock price trends. This method
has employed a Rough Set Theory (RST) algorithm which extracts optimized rules from the
optimized technical indicator dataset and deploys GA to measure the extracted rules to get
improved forecasting precision and stock price return.

Again an integrated VAR-NN-GA framework is proposed in (Ao, 2006) which makes the
automatic prediction of decision process. VAR automatically starts searching for the indicators
and correlated stocks. The NN prediction from the relevant inputs was selected by the VAR
analysis. The GA manipulates the weights of each NN proto type.TDNN & GA was used
together for accurate precision of European option prices of NIFTY index (Mishra et al, 2011).

The basic applications and extension of genetic algorithms are Genetic programming and genetic
network programming.(Chen et al,2009) use genetic network programming to structure a stock
trading model which focused on the indicators. SARSA learning algorithm is used to train the
multi-agent network system.

Binoy et al (2010) presents a data mining based stock market trend anticipation framework,
which delivers very precise stock market prediction. The proposed framework is a genetic
algorithm improved decision tree-support vector machine (SVM), which can anticipate one-day-
ahead trends in stock markets. The specialty of the proposed framework lies in the utilization of
the hybrid framework which can adjust itself to the changing market conditions and in the way
that while the greater part of the endeavors at stock market trend foreseeing have approached it
as a regression issue into a classification issue, subsequently enhancing the anticipation precision
altogether. The behavior of the proposed hybrid system is assessed on the past data from the
Bombay Stock Exchange sensitive index (BSE-Sensex). The system execution is then contrasted
with that of an Artificial Neural Network (ANN) based system and a Naïve Bayes based system.
It is discovered that the pattern anticipation precision is most elevated for the hybrid system and
the genetic algorithm optimized decision tree SVM hybrid system beats both the Artificial
Neural Network and the Naïve Bayes based pattern forecast system.
Esmaeil Hadavandi et al (2010) have introduced a hybrid methodology based on Genetic Fuzzy
Systems (GFS) and artificial neural networks (ANN) for developing a stock price estimating
master system. From the beginning, they utilized Step-wise Regression Analysis (SRA) to find
elements which have most impact on stock prices. At the following stage, they isolated their
crude data into k clusters by Self-Organizing Map (SOM) neural networks. At last, all clusters
were bolstered into autonomous GFS models with the function of filter based extraction and data
base tuning. They assessed performance of the proposed approach by applying it on stock price
data accumulated from IT and Airlines parts, and contrasted the results with past stock price
prediction methods utilizing Mean Absolute Percentage Error (MAPE).

Robert K Lai et al (2009) introduced a financial time series-forecasting model for stock market
prediction in Taiwan Stock Exchange Corporation (TSEC) by cluster the fuzzy decision tree.
This predicting prototype structure a decision-making system by combining genetic algorithm,
fuzzy decision tree (FDT) and a data clustering technique based on the past data and technical
indexes. These past historic data is clustered into k sub-clusters and the K-means algorithm is
applied. To enhance the precision of the prediction model, GA is applied to get the number of
fuzzy terms for each input index in FDT. For each sub-cluster, a separate prediction prototype is
generated.

Ching Long Su et al (2010) have created a self-organized, five-layer neuro-fuzzy model to

construct the changes of stock exchange by utilizing technical indicators. The prototype
adequacy in foreseeing was assessed by a data set containing four indicators: the stochastic
oscillator (%K and %D), volume adjusted moving average (VAMA) and ease of movement
(EMV) from TAIEX (Taiwan Stock Exchange Capitalization Weighted Stock Index). A
modified moving average technique can be proposed to foresee the input set for the neuro-fuzzy
model in determining stock cost.

In (Teoh et al,2008) a hybrid fuzzy time series model is suggested to forecast stock markets
along with the rough set rule induction and the cumulative probability distribution approach.
Cumulative probability distribution approach was used to discretize the observations in training
datasets and the rough set algorithm generates the rules. At last the forecasting was done with the
rule support values from rough set algorithm.
Yang et al (2012) propose a stock price prediction technique to generate rules in fuzzy systems
using genetic algorithm which eliminates errors due to noisy data. Finally the fuzzy reasoning
approach is applied on the rule set which anticipates the stock market price trends.

Sam Mahfoud and Ganesh Mani (1996) compares the two stochastic and real valued algorithms
namely Differential Evolution (DE) and Particle Swarm Optimization (PSO). These algorithms
predict the day by day stock market prices by using the training of feed-forward neural network.

Xiao-yu Fang and Tao Bai (2009) introduced the wavelet transform along with the ACO-SVM
for predicting stock market exchange. Wavelet Transform (WT) is used like a pre-processor of
SVM to eliminate the changing factor of actual data since the dynamic nature of time series
reduces the accuracy precision. The input data sets are chosen from the daily closing prices of
Huaneng Guoji stock in shanghai market in China.

Dassanayake and Chandima Tilakarathme (2010) uses two comfortable machine learning
techniques namely Genetic Algorithm (GA) and Neural Networks which are combined together
to predict the trading signals of Sri Lankan stock market. These two techniques in combination
predict the direction of the close price of nest day.

Karazmodeh et al (2013) uses an improved Particle Swarm Optimization (PSO) through Genetic
Algorithm (IPSO) for prediction of various stock exchanges based on Support Vector Machines
(SVM).Variety of indicators from the technical analysis field is used as input characteristics. The
correlation between stock prices of different companies has been used to forecast the price of a
stock.
Chiu and Chian (2010) investigates the dynamics of stock market from different countries in
which it combines a genetic algorithm along with support vector regression. A genetic algorithm
selects the inputs for the support vector regression.They found that the dynamics of stock
markets from Singapore, Taiwan and Indonesia were easier to validate than from the USA.