Article

Article title THE MAIN APPROACHES TO THE CREATION OF THE INFORMATION SYSTEM FOR MODELING OF TELECOMMUNICATION CLIENTS OUTFLOW
Authors S.S. Alkhasov, A.N. Tselykh
Section SECTION II. COMPUTER ENGINEERING AND COMPUTER SCIENCE
Month, Year 02, 2015 @en
Index UDC 004.89
DOI
Abstract In the present article most important functional modules of the information system for the prediction of clients outflow from a telecom company are briefly considered. The basic approaches to pre-processing of archived data and clients outflow modeling are defined. The main requirements for the practical implementation of the prognostic system are introduced. Special attention is focused on over-coming of the strong correlation between the variables in the array of input data. It"s offerred to use principal components method, implying the decomposition of the input array to the score and the loading vectors. Considered algorithm NIPALS has iterative character. The score vector calculated on some iteration is the corresponding principal component. The determination of the principal components of the long-range orders doesn"t have the sense typically because the values caused availability of some error in the input data. The basic criteria for definition of efficient number of principal components are specified: explained variance and normalized eigenvalue of score vector. As an example experimental array of the input data (9х2000) was formed. It contains selected diverse variables (technology of connection, municipality type, speed of connection, cost of service, traffic for 1st month, traffic for 2nd month, traffic for 3rd month, etc.). It"s noted that these methods allow to overcome heterogeneity of input information and the strong correlation of the variables as well as to reduce the dimensionality of the input array. It"s shown graphically how the number of used principal components influences on the explained variance and the normalized eigenvalue. All these aspects show that this approach is enough promising for use in the prognostic system containing modules of clustering and neurocomputing.

Download PDF

Keywords Prediction; clients outflow; Internet; principal component analysis; reduction of dimensionality; clustering.
References 1. Alkhasov S.S., Tselykh A.N. Printsipy postroeniya prognosticheskoy sistemy dlya modelirovaniya ottoka klientov uslug Interneta [Principles of forecasting systems for modeling churn Internet service], Materialy Vserossiyskoy nauchnoy konferentsii «Sistemy i modeli v informatsionnuyu epohu» [Materials of all-Russian scientific conference "Systems and models in the information age"]. Part 1. – Taganrog: TIT SFedU, 2014, pp. 4-6.
2. Guazzelli A. Predicting the future, Part 2: Predictive modeling techniques [Electronic resource]. Available at: http://www.ibm.com/developerworks/library/ba-predictive-analytics2 /index.html, free (Accessed: 24 November 2014).
3. Efimov A. Uderzhanie klienta – ne iskusstvo, a prikladnaya nauka! [Customer retention is not an art and applied science!] [Electronic resource]. Available at: http://library.croc.ru/ download/ 11342/930706a746182066f55bfe8f5800b735.pdf, free (Accessed: 22 November 2014).
4. Farakhutdinov Sh.F., Bushuev A.S. Obrabotka i analiz dannyih sotsiologicheskih issledovaniy v pakete SPSS 17.0 [Processing and analysis of data of sociological surveys in SPSS 17.0]. Tyumen: TyumGNGU, 2011, 220 p.
5. Guazzelli A. Predicting the future, Part 3: Create a predictive solution [Electronic resource]. Available at: http://www.ibm.com/developerworks/library/ba-predictive-analytics3/index.html, free (Accessed: 26 November 2014).
6. Baranov S.V., Skufina T.P. Novye metodiki i rezultaty issledovaniya mezhregionalnoy differentsiatsii na osnove metoda glavnyh komponent [New methods and research results of regional differentiation on the basis of principal component], Vestnik Murmanskogo
gosudarstvennogo tehnicheskogo universiteta [Bulletin of the Murmansk State Technical University], 2008, Vol. 11, No. 2, pp. 201-210.
7. Mokeev V.V., Tomilov S.V. Ob effektivnosti analiza i raspoznavaniya izobrazheniy metodom glavnyh komponent i lineynyim diskriminantnym analizom [About the effectiveness of the analysis and recognition of images by the method of principal component and linear discriminant analysis], Vestnik Yuzhno-Uralskogo gosudarstvennogo universiteta. Seriya: Kompyuternyie tehnologii, upravlenie, radioelektronika [Bulletin of the South Ural State University. Computer Technologies, Automatic Control & Radioelectronics], 2013, Vol. 13, No. 3, pp. 61-70.
8. Kim J.-O., Myuller Ch.U., Klekka U.R. et al. Faktornyiy, diskriminantnyy i klasternyy analiz [Factor, discriminant and cluster analysis]. Moscow: Finansyi i statistika, 1989, 215 p.
9. Vorontsov K.V. Regressionnyy analiz i metod glavnyh komponentov: Lektsiya [Regression analysis and principal components: Lecture]. Available at: http://shad.yandex.ru/lectures/ machine_learning_9.xml, free (Accessed 10 December 2014).
10. Rodionova O.E. Intellektualnyy metod obrabotki rezul’tatov mnogokanalnyh eksperimentov: Diss. dokt. fiz.-mat. nauk [Intelligent method of processing the results of multichannel experiments. Dr. phys. and math. sci. diss. Moscow, 2008, 272 p.
11. Esbensen K. Analiz mnogomernyh dannyh. Izbrannye glavy [Analysis of multidimensional data. Selected chapters]. Chernogolovka: IPHF RAS, 2005, 160 p.
12. Jolliffe I.T. Principal Component Analysis. 2nd ed. NY; Berlin; Heidelberg: Springer Verlag, 2002, 518 p.
13. Ovsyannikov V.P., Fefelov V.V. Opredelenie kolichestva glavnyh komponent dlya dannyh, harakterizuyuschih rabotu ugol’nogo predpriyatiya, na osnovanii analiza polnoy i ob'’yasnennoy dispersii ostatkov [The determination of the number of principal components
for the data characterizing the operation of a coal mine, on the basis of detailed analysis and explained variance of the residues], Sovremennye problemy upravleniya proizvodstvom: tezisy dokladov V Mezhdunarodnoy nauchno-prakticheskoy konferentsii [Modern problems of production management: abstracts of the V International scientific-practical conference] (Donetsk City, 14–15 October 2010). Donetsk: DonNTU, 2010, pp. 161-163.
14. Pomerantsev A. Metod glavnyh komponent (PCA) [Principal component method (PCA)]. Available at: http://www.chemometrics.ru/materials/textbooks/pca.htm, free (Accessed 09 December 2014).
15. Ayvazyan S.A., Buhshtaber V.M., Enyukov I.S., Meshalkin L.D. Prikladnaya statistika: Klassifikatsiya i snizhenie razmernosti [Applied statistics: Classification and reduction of dimensionality]. Moscow: Finansyi i statistika, 1989, 607 p.
16. Realizatsiya metoda glavnyh komponent na C# [The implementation of the method of principal components in C#]. Available at:http://habrahabr.ru/post/146236/, free. (Access: 04 December 2014).
17. Barsegyan A.A., Kupriyanov M.S., Stepanenko V.V., Kholod I.I. Tehnologii analiza dannyih: Data Mining, Visual Mining, Text Mining, OLAP [Technology data analysis: Data Mining, Visual Mining, Text Mining, OLAP]. 2nd ed. St. Petersburg: BHV-Peterburg, 2007, 384 p.
18. Medvedev V.S., Potemkin V.G. Neyronnye seti. MATLAB 6 [The neural network. MATLAB 6]. Moscow: DIALOG-MIFI, 2002, 496 p.
19. Vorontsov K.V. Nelineynaya regressiya. Mnogosloynye neyronnye seti: Lektsiya [Multilayer neural networks: Lecture]. Available at: ttp://shad.yandex.ru/lectures/machine_ learning_10.xml, free (Access: 24 November 2014).
20. Neyronnye seti. STATISTICA Neural Networks: Metodologiya i tehnologii sovremennogo analiza dannyh [The neural network. STATISTICA Neural Networks: Methodology and technology of modern data analysis]. 2nd ed. Moscow: Goryachaya liniya – Telekom, 2008, 392 p.

Comments are closed.