Dissertation > Excellent graduate degree dissertation topics show

Research and Applications of Classification Algorithms in Imbalanced Data Sets

Author: GaoJiaWei
Tutor: LiangJiYe
School: Shanxi University
Course: Systems Engineering
Keywords: Unbalanced data sets Classification ROC curve Information Granulation KAIG Fuzzy ART Telecom Customer Churn
CLC: TP301.6
Type: Master's thesis
Year: 2008
Downloads: 188
Quote: 1
Read: Download Dissertation

Abstract


Unbalanced data set is the same data set the number of samples of some classes is far less than the number of samples of other classes, is widely found in real life. Classification using traditional machine learning methods, for a small number of classes for classification accuracy rate is very low, and the number of classes is relatively high but concentrated in the non-equilibrium data, a small number of classes is often the object of concern, so the limited capacity of the traditional algorithm for solving the problem of unbalanced data sets classification. recent years classification of unbalanced data set has been widespread concern from domestic and foreign experts, has yielded some results, and has been applied in the relevant fields. In this paper, under the framework of the KAIG model point of view, based on information granules, unbalanced data sets classification problem do further research and access to knowledge, and used in the telecommunications customer churn prediction field. mainly achieved the following research results: (1 partial improvement,) on KAIG algorithm. introduced Purity parameters to measure the degree of grain overlap examples show that it is conducive to determine grain degree of overlap and can set a threshold to determine whether the degree of overlap to achieve some kind of acceptable grain tablets overlap can not be completely eliminated, this for the original KAIG model provides a new measurement tool In addition, to resolve the problem of overlapping tablets in the use of sub-attributes, the attribute value continuous data is transformed into discrete data and reuse attributes to reduce the degree of overlap of grain constantly Purity parameters to determine whether correction attribute range. Although we can not completely eliminate the grain overlap, but you can greatly reduce the tablets of the degree of overlap, to help more effectively in the property value of continuous numeric data extraction rules. The experiments show improved KAIG algorithms not only unbalanced data set classification performance compared Good, but also the classification performance of the balanced data set with other traditional classification algorithm very basic, in particular, when the attribute value is continuous values ??better than original KAIG algorithm classification performance (2) will be improved KAIG algorithm applied to telecommunications customers churn prediction. telecom customer churn is more typical unbalanced data sets, Shanxi Province, a city of a telecom operator of fixed telephone subscribers from April 2007 to July data for the training set, its extraction rules and forecast churn situation in August 2007 at the same time with the operators to C5.0 and logistic regression method as the core customer churn prediction model comparison experiments proved the effectiveness of the algorithm in telecom customer churn prediction the practical problems, the first time the ROC curve is introduced to measure the telecommunications customer churn prediction accuracy set of unbalanced data classification problem and telecom customer churn prediction problem, a number of studies, however, how the qualitative attributes or The mixed properties unbalanced datasets efficiently classification and competitor analysis, and quality of service into telecommunications customer churn prediction model also worthy of study of this research work is just a try, and related work remains to be further studied.

Related Dissertations

  1. K Company’s Improving Planning and Forecasting for the Reasonable Allocation of Inventory,F224
  2. The Simulation and Analysis of Spectrum Sensing Based on Higher Order Cumulants,TN911.23
  3. Research on the Classification Based on the Reconstruction of Solder Joint,TP391.41
  4. The Research and Implementation of Protein Classification Algorithm on the Basic of String Kernel,TP301.6
  5. Context-Dependent Lexical Paraphrasing,TP391.1
  6. Research on Text Classification Based on Biomimetic Pattern Recongnition,TP391.1
  7. Tourism Comments on the Internet’s Semantic Analysis and Usefulness Research,TP391.1
  8. Research on Classification Method of Tongue Substance Color and Tongue Coating Color Based on SVM,TP391.41
  9. Research and Application of Diverse Density Learning Algorithm,TP181
  10. Research on Target Tracking and System in Wireless Sensor Networks Based on Character of Movement and Terrain Restriction,TP212.9
  11. Study on the Preparation of the Low Molecular Weight Fuciodan from Sargassum Henslowianum (C.Agardh) and Anti-tumor Activity,TS254.9
  12. Uptake and Accumulation of Heavy Metals of Pistia Stratiotes and Eicharnia Crassipes under Combinated Pollution Condition,X173
  13. Application of Numerical Designing Cigarette Blending Formula on Computational Intelligence,TS44
  14. Establishment of the Molecular Identifying System of Curvualria and Application in Difficult Species,Q949.32
  15. Research on Nondestructive Detection Technology for External Qualities of Papayas Based-on Vision,S667.9
  16. Research on Design and Practice of Educational Video Co-Construction and Sharing Platform for the Public,G434
  17. Studies on Effects of Soybean Species on North Tofu Quality and Methods of Evaluation in North Tofu,TS214.2
  18. Comparison of Gene Expression Data Cluster Methods and Gene Network Construction for Phytophthora Sojae Genes,S435.651
  19. Study on Growth Monitoring Technique Based on Pixel Un-Mixing Method and HJ Remote Sensing Images in Paddy Rice,S511
  20. Methodology for Linkage-map-based Quantitative Trait Loci Synthesis Analysis,S562
  21. Shanghai Public Security Authorities of the Police Job Classification,D631.1

CLC: > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > General issues > Theories, methods > Algorithm Theory
© 2012 www.DissertationTopic.Net  Mobile