Year: 2013
Type: editorial
Abstract: This issue of the IDA journal consists of nine articles which represent a variety of topics, all related to the applied and theoretical research in the field of Intelligent Data Analysis.The first three articles of this issue are about semi-supervised learning methods.Imani and Keyvanpour in the first article of this issue discuss the difficulty in text categorization where a small number of examples are available and what role semi-supervised learning methods can play.They propose use of self-learning as a solution to this class of problems and apply dynamic weighting combined with majority voting approach to analyse un-labeled data.They used their proposed method on real world data sets where their experiments indicate performance improvements in classification.Choi et al. in the next article of this issue discuss the difficulty in validation of discovered models when a small number of labeled examples are available from which a subset has to be used for validation.They propose to use ensemble learning and graph sharpening to circumvent this problem.Their experimental results demonstrate the applicability of their method in real world situations where best parameter values are identified.Sugiyama and Yamamoto in the last article of this group propose a new approach for semisupervised learning that is based on closed set lattices.They present a learning algorithm which performs as multi-class classifier and a label ranker for mixed data that contains both discrete and continuous variables.Their proposed algorithm uses both labeled and unlabeled data to construct a closed set lattice.Their experiments show the competitive performance of their algorithm in classification and ranking compared to other learning algorithms.Chen et al. in the fourth article of this issue discuss the problem of skewed class distribution and costs associated with non-uniform misclassification.They investigate the effects of cost ratio, imbalance ratio and sample size on classification performance using some real-world data sets.They further demonstrate that cost ratio and level of class imbalance have strong effects on prediction performance and mostly recommend near balanced training data sets to be used.Albertini and de Mello discuss the main complexity in analyzing data streams where clustering is applied dynamically to understand and represent data behavior changes.They emphasize the limitations of the above process and propose an on-line and adaptive approach to monitor and react to behavior changes.Their proposed approach that is based on k-means is experimentally evaluated and behavior changes are verified by testing the isomorphism of Markov chains over time.This issue also includes four reviews and applied research articles.Khatoon et al. in the sixth article emphasize the importance of mining software source code for identifying useful patterns and provide a survey of tools and techniques which are based on data mining approaches.Their article provides a comparison and evaluation of some current code mining tools and techniques and presents a concise overview of source code mining techniques.Their results show that existing techniques are primarily targeting bug detection and among some of the areas where they identify a need is tools to develop quality software by automatically detecting different bugs.Covões et al. in the next article compare a number of clustering algorithms that are suitable to work with constraints.
Source: Intelligent Data Analysis
DOI: 10.3233/ida-130583
Author: A. Famili
Language: Spanish
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4245187962
