International Journal of Business, Humanities and Technology Vol. 3 No.

3; March 2013

Stock Price Prediction Using K-Nearest Neighbor (kNN) Algorithm

Khalid Alkhatib1
Hassan Najadat2
Ismail Hmeidi 3
Mohammed K. Ali Shatnawi 4

Abstract
Stock prices prediction is interesting and challenging research topic. Developed countries' economies are
measured according to their power economy. Currently, stock markets are considered to be an illustrious trading
field because in many cases it gives easy profits with low risk rate of return. Stock market with its huge and
dynamic information sources is considered as a suitable environment for data mining and business researchers.
In this paper, we applied k-nearest neighbor algorithm and non-linear regression approach in order to predict
stock prices for a sample of six major companies listed on the Jordanian stock exchange to assist investors,
management, decision makers, and users in making correct and informed investments decisions. According to the
results, the kNN algorithm is robust with small error ratio; consequently the results were rational and also
reasonable. In addition, depending on the actual stock prices data; the prediction results were close and almost
parallel to actual stock prices.
Keywords: stock price prediction, listed companies, data mining, k-nearest neighbor, non-linear regression.
1. Introduction
Recent business research interests concentrated on areas of future predictions of stock prices movements which
make it challenging and demanding. Researchers, business communities, and interested users who assume that
future occurrence depends on present and past data, are keen to identify the stock price prediction of movements
in stock markets (Kim, 2003). However, financial data is considered as complex data to forecast and or predict.
Predicting market prices are seen as problematical, and as explained in the efficient market hypotheses (EMH)
that was put forward by Fama (1990). The EMH is considered as bridging the gap between financial information
and the financial market; it also affirms that the fluctuations in prices are only a result of newly available
information; and that all available information reflected in market prices. The EMH assert that stocks are at all
times in equilibrium and are difficult for inventors to speculate. Furthermore, it has been affirmed that stock
prices do not pursue a random walk and stock prediction needs more evidence (Gallagher and Taylor, 2002;
Walczack, 2001; Kavusssanos and Dockery, 2001; Lakonishok et.al, 1994; O'Connor et. al., 1997; Lo and
MacKinlay, 1997; Kirt and Malaikah, 1992; Lo and MacKinlay, 1988). Moreover, various studies were performed
to determine stock price predictions (Subha and Nambi, 2012; Qian and Rasheed, 2007; Fama and French, 1992;
Cochrane, 1988; Campbell, 1987; Chen, et al. 1986; Basu, 1977).
In addition to purchasing and selling stocks and shares in stock markets, each stock is not only characterized by
its price, but also by other variables such as closing price which represents the most important variable for
predicting next day price for a specific stock. There is a relationship and specific behavior exists between all
variables that effect stock movements overtime. Different economic factors, such as political stability, and other
unforeseeable circumstances are variables that have been considered for stock price predictions (Ou, P. and Wang,
H., 2009; Fama and French, 1993; Cochrane, 1988; Campel, 1987; Chen. et.al.1986). Table 1 summarizes the
main variables that affect stock movements used in this article.
_________________
1, 3, 4
Department of Computer Information Systems, Jordan University of Science and Technology, Irbid, Jordan
2
Department of Computer Science, College of Computers and Information Technology, Taif University Taif,
KSA, and currently on sabbatical leave from Jordan University of science and Technology, Jordan
32
© Centre for Promoting Ideas, USA www.ijbhtnet.com

Data mining technology is used in analyzing large volume of business and financial data, and it is applied in order
to determine stock movements. Mining temporal stock markets is required to provide additional capabilities
required in cases where the existing data and their interactions need to be observed through time dimension.
In stock predictions, a set of pure technical data, fundamental data, and derived data are used in prediction of
future values of stocks. The pure technical data is based on previous stock data while the fundamental data
represents the companies’ activity and the situation of market. Combining data mining classification approaches
in stock prediction yields a future value for each unknown entities of companies’ stocks values based on historical
data. This prediction uses various methods of classification approaches such as neural networks, regression,
genetic algorithm, decision tree induction, and k-Nearest Neighbors (kNN). In classification approaches, a data set
is divided into training data set and testing set. kNN uses similarity metrics to compare a given test entity with the
training data set. Each data entity represents a record with n features. In order to predict a class label for unknown
record, kNN selects k recodes of training data set that are closest to the unknown records.
The rest of the article will be structured as follows; section 2 will exemplify the review of the relevant literature,
while section 3 describes the research methodology used and analysis. Section 4 shows the data description,
results and analysis. The Non-linear regression results are included in section 5. Finally the conclusion is seen in
section 6.
2. Literature Review
Financial services companies are developing their products to serve future prediction. There are a large amount of
financial information sources in the world that can be valuable research areas, one of these areas is stock
prediction and also called stock market mining. Stock prediction becomes increasingly important especially if
number of rules could be created to help making better investment decisions in different stock markets.
The genetic algorithm had been adopted by Shin et al. (2005); the number of trading rules was generated for
Korea Stock Price Index 200 (KOSPI 200), in Sweden Hellestrom and Homlstrom (1998) used a statistical
analysis based on a modified kNN to determine where correlated areas fall in the input space to improve the
performance of prediction for the period 1987-1996. Both models mentioned were provided in the Zimbabwe
stock exchange to predict the stock prices which included Weightless Neural Network (WNN) model and single
exponential smoothing (SES) model Mpofu (2004). Clustering stocks approach was provided by Gavrilov et al.
(2004) to group 500 stocks from the Standard & Poor. The data represented a series of 252 numbers including the
opening stock price. A fuzzy genetic algorithm was presented by Cao (1977) to discover pair relationship in stock
data based on user preferences. The study developed potential guidelines to mine pairs of stocks, stock-trading
rules, and markets; it also showed that such approach is useful for real trading. Moreover, other studies adopted
kNN as prediction techniques such as (Subha et al., 2012; Liao et al. 2010; Tsai and Hsiao 2010; Qian and
Rasheed, 2007)
3. Research Methodology And Analysis
The kNN algorithm method is used on the stock data. Also, mathematical calculations and visualization models
are provided and discussed below.
3.1 k-Nearest Neighbor Classifier (kNN)
K-nearest neighbor technique is a machine learning algorithm that is considered as simple to implement (Aha et
al. 1991). The stock prediction problem can be mapped into a similarity based classification. The historical stock
data and the test data is mapped into a set of vectors. Each vector represents N dimension for each stock features.
Then, a similarity metric such as Euclidean distance is computed to take a decision. In this section, a description
of kNN is provided. kNN is considered a lazy learning that does not build a model or function previously, but
yields the closest k records of the training data set that have the highest similarity to the test (i.e. query record).
Then, a majority vote is performed among the selected k records to determine the class label and then assigned it
to the query record.
The prediction of stock market closing price is computed using kNN as follows:
a) Determine the number of nearest neighbors, k.
b) Compute the distance between the training samples and the query record.

33
International Journal of Business, Humanities and Technology Vol. 3 No. 3; March 2013

c) Sort all training records according to the distance values.

d) Use a majority vote for the class labels of k nearest neighbors, and assign it as a prediction value of the
query record.
3.2 Mathematical Calculations and Visualizations Models
This represents an overview of equations that were applied in this article for predicting next day price. The
calculations includes error estimation, total sum of squared error, average error, cumulative closing price when
sorted using predicted values, k-values and training Root Mean Square (RMS) errors.
a) Root Mean Square Deviation (RMSD) is accuracy metric that computes the differences between the
estimated values, Y, and the actual values, X. The total of RMSD is aggregated into a single value
measure. RMSD = SQRT(Y-X)2.
b) Explained Sum of Squares (ESS) is computed as follows:

ESS =

Where yi: is the predicted variable, and y is the actual value.

c) Average Estimated Error (AEE)
AEE is the total sum of RMS errors for all variables in stock records divided by the total number of the
records.

AEE =

3.3 Visualization Graph

To evaluate the performance of kNN learning model, lift graph is applied and drawn for different companies’
stock values. The lift chart symbolizes the enhancement that a data mining model offers when distinguished
against a random estimation, and the change is expressed in terms of lift score. Through contrasting the lift scores
for a variety of parts of the data set and for different models, it can then be decided which model is supreme and
which percentage of the cases within the data set would gain from employing the predictions model. Furthermore,
using the lift chart assist in distinguishing how accurate predictions are for various models with identical
predictable characteristic. The lift graph also shows the ratio between the results obtained using the predictive
model or not. The other graph applied is the plot curves to show the relation between the actual and predicted
stock price.
4. Data Description, Results, And Analysis
In this article, data from the Jordanian stock exchange was analyzed and a brief data analysis is presented to
provide the reader with the fundamental concepts of data attributes. Also, the obtained results of prediction of the
Jordanian stock exchange are provided.
4.1 Data Description
The sample data was extracted from the Jordanian stock exchange. The study sample included stock data of five
randomly selected companies listed on the Jordanian stock exchange as a sample training dataset from the period
June 4, 2009 to December 24, 2009 as shown in table 2. Each of these companies has approximately 200 records
with three attributes including closing price, low price, and high price as shown in table 3. A brief data analysis is
presented with the fundamental concepts of data attributes. The attributes for each company are included in the
data analysis. Closing price is the main factor that affects the prediction process for a specific stock based on kNN
algorithm. The kNN algorithm is applied on a 1000 records to estimate predicted values for each stock.
4.2 Analysis And Results
The results of the predicted stock price for each individual company used in the sample with graphs for the actual
and predicted prices are presented.
The results as seen in tables 4.1 to 4.5 and in figures 1 to 11 are those after applying kNN algorithm for each
company's closing prices with the residual values which indicates how far away is the predicted values from the
actual values; the negative residual value indicates that the predicted value is larger than the actual one.
34
© Centre for Promoting Ideas, USA www.ijbhtnet.com

Section 4.3 summarizes the five companies’ prediction performance. Tables 4.1 to 4.5 respectively represent the
results after applying kNN algorithm of the Arab international for education and investment (AIEI), Jordan steel
company (JOST), Arab financial investment (AFIN), Irbid district electricity (IREL), and the Arab potash
company (APOT).
As depicted in the figures (1-10) below, the line chart of the actual and predicted values for the companies in the
sample and after adopting the kNN prediction model, the results show that the predictive value and the actual
value were moving in similar manner as seen in figure 2,4,6,8, and 10. Moreover, the lift chart also applied to
evaluate the performance of kNN learning model used and proved that the model used is performing well; this can
also be seen in figures 1,3,5,7, and 9 representing the lift charts (company's dataset).
4.3 Prediction Performance Evaluations
Table 6 represents a summary of the total squared errors, RMS errors and the average errors for the five
companies. The residuals offer the differences between the predicted values and actual the values in the sample
data. The table also shows that the values of errors are very small which indicate that the actual value and
predicted value are close. This yields a high accuracy of using the kNN algorithm in predicting stock values.
5. Non-Linear Regression Results
Non-linear regression is a data analysis technique in which the observed data is incorporated into a model
presented in a mathematical non-linear function combining the model parameters that relies on independent
variables used. GraphPad Prism v5.02 software was used to apply centered second order polynomial (quadratic)
non-linear regression which has the following formula:
Price = B0 + B1 (day – mean (day)) + B2 (day – mean (day))2
Where:
B0, B1 and B2: Constants.
Day: Actual day in which we will predict the price.
Price: Predicted price depending on the day.
Figure 11 provides a graphical representation of non-linear regression. Also, the figure shows a computed
regression equation for each company. These companies are AIEI, JOST, APOT, AFIN, and IREL. Based on
these equations, investors can compute the future values of stock prices for each company.
6. Conclusion
In this paper, a prediction process for five listed companies on the Jordanian Stock Market was carried out, and is
considered to be the first of its type implemented in Jordan as a case study using real data and market
circumstances. Consequently, a robust model was constructed for the purpose set out. The data was extracted
from five major listed companies on the Jordanian stock exchange, the sample data was used to be our training
data set (about 200 records for each company) upon the criteria previously mentioned to apply our model. We
adopted an efficient prediction algorithm tool of kNN with k=5 to perform such tests on the training data sets we
had. According to the results, kNN algorithm was stable and robust with small error ratio, so the results were
rational and reasonable. In addition, depending on the actual stock prices data; the prediction results were close to
actual prices. Having such rational results for predictions in specific, and for using data mining techniques in real
life; this presents a good indication that the use of data mining techniques could help decision makers at various
levels when using kNN for data analysis. So, we consider that employing this prediction model, kNN is real and
viable for stock predictions. Nevertheless, the implementation and the full make use of information systems
technology and due to the lack of knowledge of financial econometrics in Jordan, there is still long way to utilize
advanced predicting models to help the financial markets and brokerage houses and to move forward and be part
of the developed international financial markets. Furthermore, this may weakens the attractiveness of investments
in the Jordanian market which eventually weakens the market return. The study also shows that contemporary
data mining techniques offer the world of finance useful stock market movements' prediction analysis.

35
International Journal of Business, Humanities and Technology Vol. 3 No. 3; March 2013

References
Aha, D., Kibler, D.W., Albert, M.K. (1991). Instance-based learning algorithms. Mach Learn, 6, 37–66
Alexander, S.S. (1961). Price movements in speculative markets: Trends or random walks. Ind Manage Rev, 7–26
Breiman, L. (1996). Bagging predictors. Mach Learn 24(2), 123–140
Breiman, L. Friedman, J. Stone, C.J. Olshen, R.A. (1984). Classification and regression trees. Chapman & Hall
(Wadsworth, Inc.). NewYork
Cao, L. (1997). Practical method for determining the minimum embedding dimension of a scalar time series. Physica D
110, 43–50
Cootner, P.H. (1964.) The random character of stock market prices. MIT Press, MA
Corazza, M. Malliaris, A.G. (2002). Multi-fractality in foreign currency markets. Multinat Fin J 6(2), 65–98
Dietterich, T.G. (1997). Machine-learning research: Four current direction. AI Magazine 18(4), 97–136
Dietterich, T.G. (2000). Ensemble methods in machine learning. First International Workshop on Multiple Classifier
Systems. New York
Fama, E.F. (1965). The behaviour of stock market prices. J. Bus 38, 34–105
Fama, E.F. (1991). Efficient capital markets. II J. Fin. 46(5), 1575–1617
Fama, E.F., Fisher, L., Jensen, M., Roll, R. (1969). The adjustment of stock price to new information. Int. Eco Rev
10(1), 1–21
Frank, R.J., Davey, N. Hunt, S.P. (2000). Input window size and neural network predictors. IEEE-INNS-ENNS Int
Joint Conf Neural Netw (IJCNN’00). 2, 2237–2242
Gallagher, L. Taylor, M. (2002). Permanent and temporary components of stock prices: Evidence from assessing
macroeconomic stocks. Southern Eco J 69, 245–262
Gately, E. (1996). Neural networks for financial forecasting. Wiley, New York
Grech, D. Mazur, Z. (2004). Can one make any crash prediction in finance using the local Hurst exponent idea?
Physica A. Statistical Mech Appl. 336, 133–145
Hagan, M.T., Demuth, H.B., Beale, M.H. (1996). Neural network design. PWS Publishing, Boston, MA
Hagan, M.T., Menhaj, M. (1994). Training feedforward networks with the Marquardt algorithm. IEEE Trans Neural
Netw. 5(6), 989–993
Hansen, L. Salamon, P. (1990). Neural network ensembles. IEEE Trans Patt Analy Mach Intell. 12, 993–1001
Hellstrom, T., Holmstrom, K. (1998). Predicting the stock market. Technical report series IMa-TOM-1997-07. Center
of Mathematical Modeling, Malardalen University
Hornik, K., Stinchcombe, M., White,H. (1989). Multilayer feed forward networks are universal approximators. Neural
Net 2(5), 259–366
Hsieh, D.A. (1991). Chaos and nonlinear dynamics: application to financial markets. J Fin 46, 1839–1877
Hurst, H.E. (1951). Long-term storage of reservoirs: an experimental study. Trans Amer Soc Civil Engi 116, 770–799
Jensen, M.C. (1978). Some anomalous evidence regarding market efficiency. J Fin Eco 6, 95–102
Kavussanos, M.G., Dockery, E. (2001). A multivariate test for stock market efficiency: The case of ASE Applied
Financial Economics. 11(5), 573–579(7)
Kirt, C.B., Malaikah, S.J. (1992). Efficiency and inefficiency in thinly traded stock markets: Kuwait and Saudi Arabia.
J Bank & Fin 16 (1), 197–210
Lo, A.W., MacKinlay, A.C. (1997). Stock market prices do not follow random walks. Market Efficiency: Stock Market
Behavior in Theory and Practice 1, 363–389
Mandelbrot, B. (1982). The fractal geometry of nature. W.H.Freeman. New York
Mandelbrot, B.B., Ness, J.V. (1968). Fractional brownian motions, fractional noises and applications. SIAM Rev 10,
422–437
May, C.T. (1999). Nonlinear pricing: theory & applications. Wiley, New York
Peters, E.E. (1991). Chaos and order in the capital markets: a new view of cycles, prices, and market volatility. Wiley,
New York
Peters, E.E. (1994). Fractal market analysis: applying chaos theory to investment and economics. Wiley, New York
Qian, B., Rasheed, K. (2004). Hurst exponent and financial market predictability. Proceedings of the 2nd IASTED
international conference on financial engineering and applications. Cambridge, MA, USA, 203–209
Soofi, A.S., Cao, L. (2002). Modelling and forecasting financial data: techniques of nonlinear dynamics. Kluwer
Academic Publishers: Norwell, Massachusetts
Walczak, S. (2001). An empirical analysis of data requirements for financial forecasting with neural networks. J Manag
Infor Syst 17,4, 203–222

36
© Centre for Promoting Ideas, USA www.ijbhtnet.com

Table 1: Stock market variables that affect investor decisions in buy/sell transactions

Variable Description
Price Current price
Opening Price Opening price for a specific trading day
Closing price Closing price for a specific trading day
High Highest price in a specific day
Low Lowest price in a specific day

Table 2: Companies listed on the Jordanian stock market used in the sample

Company Symbol Description Number of records

The Arab International For Education &
AIEI 200
Investment.
JOST Jordan Steel company 200
AFIN Arab Financial Investment 181
IREL Irbid District Electricity 189
APOT The Arab Potash 200

Table 3: Variables used

Variable Name Description
Closing price Current price for a stock
Low price Lowest price in a specific day for a stock
High price Highest price in a specific day for a stock

Table 4.1: The results after applying kNN algorithm for the (AIEI)
Predicted Actual Predicted Actual Predicted Actual
Residual Residual Residual
Value Value Value Value Value Value
2.85 2.9 0.05 2.624 2.63 0.006 2.505 2.55 0.045
2.825 2.8 -0.025 2.64 2.64 0 2.505 2.5 -0.005
2.825 2.85 0.025 2.69 2.69 0 2.505 2.5 -0.005
2.9 2.9 0 2.703333 2.68 -0.023333 2.505 2.5 -0.005
2.9 2.9 0 2.568 2.62 0.052 2.505 2.5 -0.005
2.993333 3.03 0.036667 2.505 2.5 -0.005 2.505 2.5 -0.005
3 3 0 2.443333 2.4 -0.043333 2.505 2.5 -0.005
2.91 2.91 0 2.32 2.32 0 2.576364 2.55 -0.026364
2.703333 2.78 0.076667 2.42 2.42 0 2.505 2.5 -0.005
2.703333 2.65 -0.053333 2.45 2.45 0 2.576364 2.55 -0.026364
2.576364 2.67 0.093636 2.568 2.54 -0.028 2.576364 2.55 -0.026364
2.6 2.6 0 2.443333 2.48 0.036667 2.593333 2.59 -0.003333
2.505 2.5 -0.005 2.45 2.45 0 2.576364 2.55 -0.026364
2.555 2.55 -0.005 2.443333 2.45 0.006667 2.505 2.5 -0.005
2.555 2.6 0.045 2.505 2.5 -0.005 2.576364 2.56 -0.016364
2.576364 2.59 0.013636 2.568 2.54 -0.028 2.505 2.55 0.045
2.61 2.61 0 2.568 2.54 -0.028 2.576364 2.55 -0.026364
2.624 2.6 -0.024 2.51 2.51 0 2.56 2.56 0
2.61 2.61 0 2.42 2.42 0 2.505 2.5 -0.005

Note: AIEI predicted closing prices after applying kNN algorithm. 200 records from the period Jan 4, 2009 to Dec
24, 2009 are selected as the training dataset and only 57 records are shown in the table. Total squared RMS error
is 0.263, RMS error is 0.0378 and the average error is -5.434E-09

37
International Journal of Business, Humanities and Technology Vol. 3 No. 3; March 2013

Table 4.2: The results after applying kNN algorithm for Jordan steel company (JOST)

Predicted Actual Predicted Actual Predicted Actual

Residual Residual Residual
Value Value Value Value Value Value
2.47 2.47 0 2.706667 2.73 0.023333 2.64 2.68 0.04
2.473333 2.47 -0.003333 2.69 2.69 0 2.59 2.59 0
2.47 2.47 0 2.703333 2.73 0.026667 2.672 2.65 -0.022
2.5325 2.52 -0.0125 2.672 2.68 0.008 2.703333 2.72 0.016667
2.615 2.58 -0.035 2.703333 2.66 -0.043333 2.755 2.7 -0.055
2.64 2.61 -0.03 2.69 2.69 0 2.82 2.82 0
2.64 2.61 -0.03 2.72 2.76 0.04 2.82 2.83 0.01
2.64 2.65 0.01 2.672 2.65 -0.022 2.962 2.9 -0.062
2.63 2.62 -0.01 2.63 2.64 0.01 2.9275 2.9 -0.0275
2.63 2.64 0.01 2.706667 2.69 -0.016667 2.923333 2.92 -0.003333
2.63 2.62 -0.01 2.79 2.79 0 2.83 2.83 0
2.645 2.65 0.005 2.672 2.71 0.038 2.8725 2.85 -0.0225
2.615 2.65 0.035 2.516667 2.59 0.073333 2.95 2.94 -0.01
2.64 2.63 -0.01 2.473333 2.47 -0.003333 2.9275 2.95 0.0225
2.64 2.7 0.06 2.516667 2.47 -0.046667 2.896667 2.9 0.003333
2.64 2.6 -0.04 2.516667 2.49 -0.026667 2.8725 2.93 0.0575
2.675 2.67 -0.005 2.5325 2.51 -0.0225 2.923333 2.88 -0.043333
2.645 2.6 -0.045 2.5325 2.51 -0.0225 2.98 2.95 -0.03

Note: 200 records from the period Jan 4, 2009 to Dec 24, 2009 are chosen as the training dataset and only 57
records are shown in the table. Total squared RMS error is 0.263, RMS error is 0.0378and the average error is -
5.4347E-09
Table 4.3: The results after applying kNN algorithm for Arab financial investment (AFIN)
Predicted Actual Predicted Actual Predicted Actual
Residual Residual Residual
Value Value Value Value Value Value
3.396667 3.39 -0.006667 3.31 3.25 -0.06 3.36 3.4 0.04
3.338 3.39 0.052 3.31 3.25 -0.06 3.384286 3.39 0.005714
3.431667 3.4 -0.031667 3.24 3.24 0 3.36 3.32 -0.04
3.396667 3.4 0.003333 3.23 3.23 0 3.37 3.43 0.06
3.431667 3.4 -0.031667 3.28 3.28 0 3.386667 3.44 0.053333
3.413333 3.4 -0.013333 3.356667 3.36 0.003333 3.384286 3.33 -0.054286
3.431667 3.4 -0.031667 3.32 3.32 0 3.4 3.4 0
3.431667 3.4 -0.031667 3.245 3.25 0.005 3.413333 3.4 -0.013333
3.49 3.46 -0.03 3.33 3.27 -0.06 3.396667 3.4 0.003333
3.49 3.46 -0.03 3.338 3.3 -0.038 3.384286 3.4 0.015714
3.49 3.46 -0.03 3.295 3.31 0.015 3.384286 3.35 -0.034286
3.338 3.4 0.062 3.356667 3.31 -0.046667 3.413333 3.44 0.026667
3.24 3.24 0 3.338 3.3 -0.038 3.465 3.45 -0.015
3.245 3.24 -0.005 3.33 3.39 0.06 3.384286 3.4 0.015714
3.24 3.24 0 3.338 3.3 -0.038 3.44 3.44 0
3.31 3.3 -0.01 3.295 3.28 -0.015 3.465 3.48 0.015

Note: AFIN predicted closing prices after applying kNN algorithm. 181 records from the period Jan 4, 2009 to
Dec 24, 2009 are chosen as the training dataset and only 54 records are shown in the table. Total squared RMS
error is 0.263, RMS error is 0.036 and the average error is -1.005E-08

38
© Centre for Promoting Ideas, USA www.ijbhtnet.com

Table 4.4 The results after applying kNN algorithm for Irbid district electricity (IREL)
Predicted Actual Predicted Actual Predicted Actual
Residual Residual Residual
Value Value Value Value Value Value
9.15 9.15 0 8.35 8.35 0 7.58 7.58 0
8.72 8.72 0 8.45 8.5 0.05 7.444 7.22 -0.224
8.803333 8.81 0.006667 8.1 8.1 0 7.574615 7.5 -0.074615
9.026667 9 -0.026667 8.17 8.17 0 7.444 7.78 0.336
9.28 9.28 0 8.6 8.6 0 7.41 7.41 0
9.2 9.2 0 8.45 8.4 -0.05 7.8 7.8 0
9.026667 9.08 0.053333 7.995 8.01 0.015 7.792 7.6 -0.192
8.803333 8.8 -0.003333 8.4 8.4 0 7.8 7.8 0
8.803333 8.8 -0.003333 8.29 8.29 0 7.792 7.6 -0.192
9.026667 9 -0.026667 8.095 7.9 -0.195 8 8 0
9.24 9.24 0 8.095 8.29 0.195 7.574615 7.74 0.165385
9.1 9.1 0 7.825 7.95 0.125 7.346667 7.38 0.033333
9.4 9.4 0 7.8 7.8 0 7.03 7.03 0
8.87 8.99 0.12 7.574615 7.6 0.025385 7.433333 7.4 -0.033333
8.87 8.75 -0.12 7.87 7.87 0 7.71 7.71 0
8.51 8.51 0 7.574615 7.5 -0.074615 7.574615 7.5 -0.074615
8.56 8.55 -0.01 7.574615 7.5 -0.074615 7.42 7.42 0
8.56 8.57 0.01 7.8 7.8 0 7.07 7.07 0

Note: AFIN predicted closing prices after applying kNN algorithm. 189 records from the period Jan 4, 2009 to
Dec 24, 2009 are chosen as the training dataset and only 54 records are shown in the table. Total squared RMS
error is 1.282, RMS error is 0.1046 and the average error is 4.2735E-08
Table 4.5: The results after applying kNN algorithm for The Arab potash (APOT)
Predicted Actual Predicted Actual Predicted Actual
Residual Residual Residual
Value Value Value Value Value Value
33.145 34.18 1.035 32.76 32.76 0 29.97 29.97 0
33.82 33.51 -0.31 34.29 34.48 0.19 30.69 30.69 0
33.77 34 0.23 34.533333 34.05 -0.483333 30.34 30.34 0
34 34 0 34.01 34.01 0 30.36 30.36 0
34.49 34.49 0 33.82 33.95 0.13 31.945 31.9 -0.045
34.98 34.98 0 33.77 33.54 -0.23 32.25 31.5 -0.75
35.166667 34.87 -0.296667 34.833333 34.25 -0.583333 33.145 32.7 -0.445
34.75 34.75 0 34.4 34.4 0 33.65 33.3 -0.35
34.99 34.99 0 33.525 33.51 -0.015 32.85 32.85 0
34.6 34.6 0 33.525 33 -0.525 33.11 33.11 0
34.29 34.1 -0.19 33.145 33 -0.145 34.61 34.61 0
34.825 34.45 -0.375 33 33 0 34.84 34.84 0
34.39 34.39 0 33.12 33.12 0 33.525 33.6 0.075
34.29 34.29 0 31.6 31.6 0 32.9 32.9 0
34.533333 34.4 -0.133333 30.88 30.88 0 31.91 31.91 0
34.3 34.3 0 29.9 29.9 0 31.945 31.99 0.045
34.435 33.97 -0.465 30.4 30.4 0 32.9 32.9 0

Note: APOT predicted closing prices after applying kNN algorithm. 200 records from the period Jan 4, 2009 to
Dec 24, 2009 are chosen as the training dataset and only 51 records are shown in the table. Total squared RMS
error is 22.7453, RMS error is 0.337 and the average error is 2.49E-08

39
International Journal of Business, Humanities and Technology Vol. 3 No. 3; March 2013

Table 6: Prediction performance evaluations for the whole sample companies (five companies) after
applying kNN algorithm for k=5

kNN algorithm for K = 5

Company Total squared RMS error RMS error Average error
AIEI 0.263151 0.0378176 -5.43E-09
AFIN 0.2629177 0.0363482 -1.01E-08
APOT 22.74533 0.3372338 2.50E-08
IREL 1.2823397 0.1046908 4.27E-08
JOST 0.17963 0.0300444 1.508E-08

Figure 1
Lift chart for AIEI training dataset

Note: The area between the baseline and the curve is an indicator of the goodness of the model.

Figure 2 : Plot graph shows the relationship between AIEI’s predicted/actual closing price for 1 year
period
3.2

2.8

Predicted Value
2.6
Actual Value

2.4

2.2

2
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55

40
© Centre for Promoting Ideas, USA www.ijbhtnet.com

Figure 3: Lift chart for JOST training data set

Note: The area between the baseline and the curve is an indicator of the goodness of the model.
Figure 4: Plot graph shows the relationship between JOST’s predicted/actual closing price for 1 year
period
3.2

2.8

Predicted Value
2.6
Actual Value

2.4

2.2

2
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55

Figure 5: Lift chart for AFIN company training data set

Note: The area between the baseline and the curve is an indicator of the goodness of the model.

41
International Journal of Business, Humanities and Technology Vol. 3 No. 3; March 2013

Figure 6: Plot graph shows the relationship between AFIN’s predicted/actual closing price for 1 year
period
3.6

3.4

Predicted Value
Actual Value

3.2

3
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55

Figure 7: Lift chart for IREL company training data set

Note: The area between the baseline and the curve is an indicator of the goodness of the model.

Figure 8: Plot graph shows the relationship between IREL’s predicted/actual closing price for 1 year
period
9.8
9.6
9.4
9.2
9
8.8
8.6
8.4
8.2
8
7.8
Predicted Value 7.6
7.4
Actual Value 7.2
7
6.8
6.6
6.4
6.2
6
5.8
5.6
5.4
5.2
5
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55

42
© Centre for Promoting Ideas, USA www.ijbhtnet.com

Figure 9 : Lift chart for APOT company training data set

Note: The area between the baseline and the curve is an indicator of the goodness of the model.

Figure 10: Plot graph shows the relationship between APOT’s predicted/actual closing price for 1 year period.

35.5

35

34.5

34

33.5

33

Predicted Value 32.5

Actual Value 32

31.5

31

30.5

30

29.5

29
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55

43
International Journal of Business, Humanities and Technology Vol. 3 No. 3; March 2013

Figure 11: Regression distributions for the sample companies

AIEI, JOST, APOT, AFIN, and IREL.

44