Optimized Back propagation Learning in Neural Networks with Bacterial Foraging Optimization to Predict Forex Gold Index (XAUUSD).

(1)

(2)

Vol. 7, no. 117-120, 2013

APPLIED MATHEMATICAL SCIENCES

Journal for Theory and Applications

(3)

Applied Mathematical Sciences

Aims and scopes: The journal publishes refereed, high quality original research papers in all branches of the applied mathematical sciences.

Call for papers: Authors are cordially invited to submit papers to the editorial office by e-mail to: ams@m-hikari.com . Manuscripts submitted to this journal will be considered for publication with the understanding that the same work has not been published and is not under consideration for publication elsewhere.

Instruction for authors: The manuscript should be prepared using LaTeX or Word processing system, basic font Roman 12pt size. The papers should be in English and typed in frames 14 x 21.6 cm (margins 3.5 cm on left and right and 4 cm on top and bottom) on A4-format white paper or American format paper. On the first page leave 7 cm space on the top for the journal's headings. The papers must have abstract, as well as subject classification and keywords. The references should be in alphabetic order and must be organized as follows:

[1] D.H. Ackley, G.E. Hinton and T.J. Sejnowski, A learning algorithm for Boltzmann machine, Cognitive Science, 9 (1985), 147-169.

[2] D.O. Hebb, Organization of Behaviour, Wiley, New York, 1949.

Editorial office

e-mail: ams@m-hikari.com

Postal address: Street address:

HIKARI Ltd, P.O. Box 15 HIKARI Ltd, Vidin str. 40, office no. 3

Ruse 7005, Bulgaria Ruse 7012, Bulgaria

www.m-hikari.com

(4)

8/10/2014 Online Edition

http://www.m-hikari.com/ams/ams-2014/ams-37-40-2014/index.html 1/2

Site Search: Go

APPLIED MATHEMATICAL SCIENCES Online Edition, Vol. 8, 2014, no. 37-40

Y. J. Lan, T. W. Hsu

Analytical solution for wave interaction with a stack-type double-layer composite poroelastic submerged structure Applied Mathematical Sciences, Vol. 8, 2014, no. 37, 1799-1816

http://dx.doi.org/10.12988/ams.2014.42132

Jelena Vujakovic

The mean value theorem of line complex integral and Sturm function

Applied Mathematical Sciences, Vol. 8, 2014, no. 37, 1817-1827 http://dx.doi.org/10.12988/ams.2014.4279

Hashem Salarzadeh Jenatabadi

An application of moderation analysis in structural equation modeling: a comparison study between MIS and ERP

Marcos Amaku, Rafael I. Cipullo, Jose H. H. Grisi-Filho, Fernando S. Marques, Raul Ossada The friendship paradox in scale-free networks

I Made Bayu Permana Putra, Rukmi Sari Hartati, I Ketut Gede Darma Putra, Ni Kadek Ayu Wirdiani

Optimized back propagation learning in neural networks with bacterial foraging optimization to predict forex gold index (XAUUSD)

Christos E. Kountzakis Credit risk transformations

Joemar C. Endam, Jocelyn P. Vilela

The second isomorphism theorem for B-algebras

Burak Kurt

On the algebraic structures of soft sets in logic

Admi Nazra

Anosov diffeomorphisms on real Bott manifolds

Ather Qayyum, Muhammad Shoaib, Muhammad Amer Latif

A generalized inequality of Ostrowski type for twice differentiable bounded mappings and applications

Jing Na, Yongkun Tang

Weak and strong convergence theorems of fixed points for total asymptotically nonexpansive multi-valued mappings in Banach spaces

Dae San Kim, Taekyun Kim, Sang-Hun Lee, Jong Jin Seo

Some symmetric identities involving Carlitz-type q-Euler polynomials

(5)

8/10/2014 Online Edition

http://www.m-hikari.com/ams/ams-2014/ams-37-40-2014/index.html 2/2

http://dx.doi.org/10.12988/ams.2014.42190

P. M. Sitharselvam, K. T. Nagalakshmi

A note on anti fuzzy sub-bi HX group and its bi level sub-bi HX groups Applied Mathematical Sciences, Vol. 8, 2014, no. 39, 1925-1932 (retracted)

S. Narayanamoorthy, A. Tamilselvi, P. Karthick, S. Kalyani, S. Maheswari Regular and totally regular bipolar fuzzy hypergraphs

G.Sh. Tsitsiashvili, Yu.N. Kharchenko, A.S. Losev, M.A. Osipova Enumeration of minimal control sets of vertices in oriented graph

Nazmiye Yilmaz, Necati Taskara

Tribonacci and Tribonacci-Lucas numbers via the determinants of special matrices

Jianchu Jiang, Xianhua Tang

Oscillation of second order half-linear difference equations (I)

A. H. Essawy, H. M. Brakat, Khloud R. Khater

Hydromagnetic self-gravitating stability of streaming fluid cylinder with longitudinal magnetic field

M. D. H. Gamal

A note on ones assignment method

Section on Mathematical Education Catarina Cosme, Cecilia Costa

On the transmission of the notion of norms having matrix values in a peripheral European country: Portugal

(6)

Applied Mathematical Sciences, Vol. 8, 2014, no. 38, 1847 - 1854 HIKARI Ltd, www.m-hikari.com

http://dx.doi.org/10.12988/ams.2014.4292

Optimized Back propagation Learning in Neural

Networks with Bacterial Foraging Optimization to

Predict Forex Gold Index (XAUUSD)

I Made Bayu Permana Putra

Department of Information Technology Udayana University, Bali, Indonesia

Rukmi Sari Hartati

Department of Electrical Engineering Udayana University, Bali, Indonesia

I Ketut Gede Darma Putra

Ni Kadek Ayu Wirdiani

Copyright ©2014 I Made Bayu Permana Putra, Rukmi Sari Hartati, I Ketut Gede Darma Putra, and Ni Kadek Ayu Wirdiani. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract

(7)

1848 I Made Bayu Permana Putra et al.

with bacterial foraging optimization. Proven by adding bacterial foraging optimization method, it involve the back propagation method can run 50% faster and accuracy increased 2% average. System performance testing was conducted 1000 times using many different data. Sample data and test data normalization is immersely affect the system performance.

Keywords : back propagation, bacterial foraging optimization, forex.

1 Introduction

International monetary market needs important data to its trade from forex rate. The index value is influenced by many factors such as economics and politics, and also traders and investors psychological condition. All of that factors make the gold index are very difficult to predict. People that involved in the field of international monetary exchanges have been searching for an explanation to improve the prediction ability. It is this ability to correctly predict gold index rate changes that allows for the maximization of profits. Trading at the right time with the relatively correct strategies can bring large profit, but a trade based on wrong movement can risk big losses. Using the right analytical tool and good methods can reduce the effect of mistakes and also can increase profitability [2][8].

Artificial neural networks has been extensively used in these days in various aspects of science and engineering because of its ability to model both linear and non-linear systems without the need to make assumptions as are implicit in most traditional statistical approaches. ANN has been a aggressive model over the simple linear regression model [1][5][9][10].Neural network algorithms can predict foreign currency exchange rate. Because of back propagation algorithm is used with AFERFM, which gives the 11.3 percentages more accurate than the HFERFM[6].

(8)

Optimized back propagation learning in neural networks 1849

2 Determination of the Architecture

Architecture determination is conducted to assign inputs number on neural, hidden layer, output layer, sample data and test data date of use. The number of input will affect to how many number of history data that will be used in back propagation process. For example if the hidden layers input which entered is 5 and the date is 2011-01-01 to 2011-01-20, then the sample data that will be formed after being normalized is as follow:

Table 1 Sample data

1 days ago 2 days ago 3 days ago 4 days ago 5 days ago today date

0.0820704 0.0936164 0.102236 0.138942 0.17167 0.088500 2011-01-10

0.0885004 0.0820704 0.0936164 0.102236 0.138942 0.091023 2011-01-11

0.0910233 0.0885004 0.0820704 0.0936164 0.102236 0.103515 2011-01-12

0.103515 0.0910233 0.0885004 0.0820704 0.0936164 0.111172 2011-01-13

0.111172 0.103515 0.0910233 0.0885004 0.0820704 0.109315 2011-01-14

3 Determination of the Bacteria Dimension

Bacterial foraging optimization will be optimized weight and bias before using back propagation method, where weights and bias optimized result eventually will be used in back propagation process. To determine cost value on bacterial foraging optimization, it can be calculated with augment feed forward process. Where dimensions of bacteria is weights and bias as on back propagation, then data which included is average value per colom on sample data.

How many dimension of the bacteri is determined in accordance to neural architectural, calculated as [3][4]:

(9)

The bacteria foraging optimization algorithm is:

1. Chemotaxis : Bacteria has a habit to move to approach a food source. This process is the way how bacteria determine its movement between dip and fall because of flagella. Bacteria will automatically move to an error optimum (nutrient source). It contains two steps as follows: [3][4]:

 Tumbling: Is the process how bacteria falls to the best nourishment position, to get minimum errors of ANN[3][4].

_√

Where

: Random vectors on [-1,1]

√ : The unit walk in the random direction C : Run length unit

: It may represent by P (i, j, k, ell)

: It may represent by P (i, j+1, k, ell) that represents

the next location of bacterial

 Swimming: A Bacteria will move towards a certain direction that has a growing number of nourishment (rich nourishment) [3][4].

2. Swarming: While bacteria is in a group, they have a censor from a dangerous place, and automatically will stay y from it by following nutrient gradient that generated by a group of bacteria that obtain nourishment [3][4].

( ) ∑

∑ ∑

Where

depth attractant to set a magnitude of secretion of attractant by a cell

width attractant to set how the chemical cohesion signal diffuse

height repellant to set repellant

 Reproduction : The fitness value from bacteria sequenced in the array sequences that order the lowest value to the highest value [3][4].

∑

Where Nc is the chemotaxis steps and j is the value of error.

(3.1)

(3.2)

(10)

 Elimination and Dispersal: In the real world bacteria has probability and spreads to a new location. Here begins by generating random vector. [3][4].

4 Bacterial Foraging Optimization Optimized Back propagation

The optimized back propagation process plot with bacterial foraging optimization is as follows:

Fig 2 A Framework of the study

Sample data is calculated, the average is used in bacteri initialization process, then the process is continued on bacterial training to detect the lowest cost value which eventually that value is the weight value and the optimal bias. After weights value and the optimal bias are found, it will be continued on back propagation process to find the weights result and bias that will be used to predict.

5 Result of Experiment

Result of experiment with sample data and data test from 2011-01-01 to 2011-12-31 and 2012-01-01 to 2012-12-31. The architecture was formed by input layer is 5, hiden layer is 10, output layer is 1. Bacterial foraging optimization process that uses chemotaxis is 5, swim step is 5, reproduction is 4, dispersal is 4, dispersal probability is 0.5, and movement size is 0.5. Back propagation learning rate process is 0.05, momentum is 0.01, max epoch is 2000, means square error target is 0.000001.

Sample data Initialization bacteria

Start BFO BFO optimizes weights

(11)

Table 2 Comparation BP with BPBFO Sample

data

Method Interaction epoch mse time

(second)

Accuration Rate

2011 BP 232460 788 0.0000008 216 92,43%

2011 BPBFO 134225 455 0.0000007 141 94,60%

2012 BP 51744 176 0.0000006 43 95,20%

2012 BPBFO 31458 107 0.0000005 26 96,26%

Table 2 shows the processing result of back propagation method and the back propagation method was developed by bacterial foraging optimization and comparation between the result accuracy from back propagation method and back propagation method that optimized with bacterial foraging optimization.

(a) (b)

Fig 3 (a) BP Result, (b) BPBFO Result test data 2011

(a) (b)

Fig 4 (a) BP Result, (b) BPBFO Result test data 2012

Fig 3 and fig 4 shows the comparison of test data and the predicted results where the red line is the target of the data test and the blue line is the predicted outcome.

6 Conclusion

(12)

results by using back propagation. Sample and data test should be tested against the data history of stable gold index is not in a state of crisis. Proved that after optimized with bacterial foraging optimization method it get significantly better in terms of iterating, epoch, and the value of accuracy. Forex gold index prediction can be developed with another neural network method to give better accuration. In further developments, this paper is expected to be a reference in the development of other forex gold index prediction to make better accurate prediction.

References

[1] T. Abe and T. Saito. An approach to prediction of spatio-temporal patterns based on binary neural networks and cellular automata. IEEE International Joint Conference on Neural Networks. 2008; 2494-2499.

[2] Abraham, Ajith and Chowdhury, Morshed U. “An Intelligent FOREX Monitoring System.” In Proceedings of IEEE International Conference on Info-tech and Infonet, Beijing, China, pp 523-528, 2001.

[4] I. AL-Hadi, Siti Zaiton. Bacterial Foraging Optimization Algorithm For Neural Network Learning Enhancement. International Journal of Innovative Computing 01(1) 8-14. 2012.

[5] O. Deperlioglu and Utku Kose. An educational tool for artificial neural networks. Computers and Electrical Engineering. 2011; 37: 392-402.

[6] R. Divyapria. 2013. Accurate Forecasting Prediction of Foreign Exchange Rate Using Neural Network Algorithms. (IJCSMC) International Journal of Computer Science and Mobile Computing. Vol. 2, Issue. 7, July 2013, pg.344 – 349

[7] D. Hwa Kim, Ajith Abraham, Jae Hoon Cho A hybrid genetic algorithm and bacterial foraging approach for global optimization, Information Sciences 177 (2007) 3918–3937

(13)

[9] S.Lee, S.Cho and P.M. Wong 1998,”Rainfall prediction using Artificial neural networks”. journal of geographic information and Decision Analysis, Vol .2,No2,pp.233-242

[10] H Zhao, J Zhang. Pipelined Chebyshev Functional Link Artificial Recurrent Neural Network for Nonlinear Adaptive Filter. IEEE Trans. Systems, Man, and Cybernetics, Part B: Cybernetics. 2010; 40:162-172.

(14)

8/10/2014 Indexing and abstracting

Applied Mathematical Sciences is indexed and abstracted by Scopus. The bibliometric indicators of Applied Mathematical Sciences have been computed by Scopus/SCImago and Journal Metrics as follows:

Cites per Doc: 0,807. Cites per Doc is computed by the same formula as journal impact factor of ISI-Thomson Reuters.

SNIP (Source Normalized Impact per Paper): 0,781 IPP (Impact per Publication): 0,673

Current Mathematical Publications (USA) Zentralblatt fur Mathematik (Germany) Directory of Open Access Journals (Sweden) Google Scholar (USA)

Academic Journals Database (Switzerland)

ROAD Directory of Open Access Scholarly Resources (EU) EBSCO (USA)

Cites per Doc Cites per Doc is a measure reflecting the average number of citations received per paper published in the journal during the two preceding years, namely,

A = the number of times articles published in the preceding two years which were cited by all journals during the current year.

B = the total number of papers published by the journal in the preceding two years. Cites per Doc = A/B.

(15)

8/10/2014 Indexing and abstracting

http://www.m-hikari.com/ams/ind.html 2/2

SNIP (Source Normalized Impact per

Paper) Created by Professor Henk Moed at CTWS, University of Leiden, Paper (SNIP) measures contextual citation impact by weighting citations based on the total number ofSource Normalized Impact per citations in a subject field.

IPP (Impact per Publication) The Impact per Publication measures the ratio of citations in a year (Y) to papers published in the three previous years (Y1, Y2, Y3) divided by the number of papers published in those same years (Y1, Y2, Y3).

SCImago journal rank indicator SCImago journal rank indicator is developed by Scopus and expresses the average number of weighted citations received in the selected year by the papers published in the selected journal in the three previous years.

H index The H index expresses the journal's number of articles (h) that have received at least h citations. It quantifies both scientific productivity and scientific impact of a journal and it is also applicable to scientists, countries, etc.

(16)

Home

Just copy the code below and paste within your html page:

How to cite this website?

Subject Category: Applied Mathematics

Publisher: Hikari Ltd.. Publication type: Journals. ISSN: 1312885X

Coverage: 1984, 2009-2013

H Index: 13

Charts Data

Indicators 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 SJR 0,000 0,000 0,000 0,000 0,000 0,000 0,000 0,000 0,000 0,000 0,000 0,136 0,270 0,235 Total Documents 0 0 0 0 0 0 0 0 0 0 260 307 351 617 Cites / Doc. (4years) 0,00 0,00 0,00 0,00 0,00 0,00 0,00 0,00 0,00 0,00 0,00 0,21 0,44 0,42 Cites / Doc. (3years) 0,00 0,00 0,00 0,00 0,00 0,00 0,00 0,00 0,00 0,00 0,00 0,21 0,44 0,42 Cites / Doc. (2years) 0,00 0,00 0,00 0,00 0,00 0,00 0,00 0,00 0,00 0,00 0,00 0,21 0,44 0,46 References / Doc. 0,00 0,00 0,00 0,00 0,00 0,00 0,00 0,00 0,00 0,00 12,47 13,33 13,74 13,91

Cited Docs. 0 0 0 0 0 0 0 0 0 0 0 41 139 223

Uncited Docs. 0 0 0 0 0 0 0 0 0 0 0 219 428 695 % International

Collaboration 0,00 0,00 0,00 0,00 0,00 0,00 0,00 0,00 0,00 0,00 6,15 6,19 8,26 9,89