Konrad Jackowski

IMPROVED ADAPTIVE SPLITTING AND SELECTION: THE HYBRID TRAINING METHOD OF A CLASSIFIER BASED ON A FEATURE SPACE PARTITIONING

Currently, methods of combined classification are the focus of intense research. A properly designed group of combined classifiers exploiting knowledge gathered in a pool of elementary classifiers can successfully outperform a single classifier. There are two essential issues to consider when creating combined classifiers: how to establish the most comprehensive pool and how to design a fusion model that allows for taking full advantage of the collected knowledge. In this work, we address the issues and propose an AdaSS+, training algorithm dedicated for the compound classifier system that effectively exploits local specialization of the elementary classifiers. An effective training procedure consists of two phases. The first phase detects the classifier competencies and adjusts the respective fusion parameters. The second phase boosts classification accuracy by elevating the degree of local specialization. The quality of the proposed algorithms are evaluated on the basis of a wide range of computer experiments that show that AdaSS+ can outperform the original method and several reference classifiers.

Algorithm of designing compound recognition system on the basis of combining classifiers with simultaneous splitting feature space into competence areas

The paper presents the novel adaptive splitting and selection algorithm (AdaSS) used for learning compound pattern recognition system. Splitting a feature space into its constituents and selection of the best area classifier from the pool of available recognizers for each region are key processes of the proposed model. Both take place simultaneously as part of a compound optimization process aimed at maximizing system performance. Evolutionary algorithms are used to find out the optimal solution. The results of experiments for algorithm evaluation purposes prove the quality of the proposed approach.

Proceedings of the 8th International Conference on Computer Recognition Systems CORES 2013

The computer recognition systems are nowadays one of the most promising directions in artificial intelligence. This book is the most comprehensive study of this field. It contains a collection of 86 carefully selected articles contributed by experts of pattern recognition. It reports on current research with respect to both methodology and applications. In particular, it includes the following sections:Biometrics Data Stream Classification and Big Data AnalyticsFeatures, learning, and classifiers Image processing and computer vision Medical applications Miscellaneous applications Pattern recognition and image processing in roboticsSpeech and word recognitionThis book is a great reference tool for scientists who deal with the problems of designing computer pattern recognition systems. Its target readers can be the as well researchers as students of computer science, artificial intelligence or robotics.

Fixed-size ensemble classifier system evolutionarily adapted to a recurring context with an unlimited pool of classifiers

This paper presents a novel ensemble classifier system designed to process data streams featuring occasional changes in their characteristics (concept drift). The ensemble is especially effective when the concepts reappear (recurring context). The system collects information on emerging contexts in a pool of elementary classifiers trained on subsequent data chunks. The pool is updated only when concept drift is detected. In contrast to other ensemble solutions, classifiers are not removed from the pool, and therefore, knowledge of past contexts is preserved for future use. To ensure high classification performance, the number of classifiers contributing to decision-making is fixed and limited. Only selected elements from the pool can join the decision-making ensemble. The process of selecting classifiers and adjusting their weights is realized by an evolutionary-based optimization algorithm that aims to minimize the system misclassification rate. Performance of the system is evaluated through a series of experiments presenting some key features of the system.

Cost-Sensitive splitting and selection method for medical decision support system

The paper presents a cost-sensitive modification of the Adaptive Splitting and Selection (AdaSS) algorithm, which trains a combined classifier based on a feature space partitioning. In this study the algorithm considers constraints put on the cost of selected features, which are one of the key-problems in the clinical decision support systems. The modified version takes into consideration both the overall classification accuracy and the cost constraints, returning balanced solution for the problem at hand. Proposed method was evaluated on the basis of computer experiments run on cost-sensitive medical benchmark datasets.

Evolutionary Algorithm for Decision Tree Induction

Decision trees are among the most popular classification algorithms due to their knowledge representation in form of decision rules which are easy for interpretation and analysis. Nonetheless, a majority of decision trees training algorithms base on greedy top-down induction strategy which has the tendency to develop too complex tree structures. Therefore, they are not able to effectively generalise knowledge gathered in learning set. In this paper we propose EVO-Tree hybrid algorithm for decision tree induction. EVO-Tree utilizes evolutionary algorithm based training procedure which processes population of possible tree structures decoded in the form of tree-like chromosomes. Training process aims at minimizing objective functions with two components: misclassification rate and tree size. We test the predictive performance of EVO-Tree using several public UCI data sets, and we compare the results with various state-of-the-art classification algorithms.

Multiple classifier system with radial basis weight function

The paper presents novel algorithm of decision making in multiple classifier system (MCS), which response is based on weighted fusion of discriminating functions derived from a pool of elementary classifiers Radial basis function model are used to establish the weights of the classifiers over a feature space For best exploitation of knowledge collected by the classifiers parameters of the weight functions are set during learning process of the MCS that aims at minimizing misclassification rate of the MCS Quality of the proposed radial basis function MCS (RB MCS) is verified in the set of experiments carried out on the set of benchmark datasets derived from UCI repository.

Application of AdaSS Ensemble Approach for Prediction of Power Plant Generator Tension

The paper presents the application of ensemble approach in the prediction of tension in a power plant generator. The proposed Adaptive Splitting and Selection (AdaSS) ensemble algorithm performs fusion of several elementary predictors and is based on the assumption that the fusion should take into account the competence of the elementary predictors. To take full advantage of complementarity of the predictors, the algorithm evaluates their local specialization, and creates a set of locally specialized predictors. System parameters are adjusted using evolutionary algorithms in the course of the learning process, which aims to minimize the mean squared error of prediction. Evaluation of the system is carried on an empirical data set and is compared to other classical ensemble methods. The results show that the proposed approach effectively returns a more consistent and accurate prediction of tension, thereby outperforming classical ensemble approaches.

Adaptive Splitting and Selection Method of Classifier Ensemble Building

The paper presents a novel machine learning method which allows obtaining compound classifier. Its idea bases on splitting feature space into separate regions and choosing the best classifier from available set of recognizers for each region. Splitting and selection take place simultaneously as a part of an optimization process. Evolutionary algorithm is used to find out the optimal solution. The quality of the proposed method is evaluated via computer experiments.

APPLICATION OF ADAPTIVE SPLITTING AND SELECTION CLASSIFIER TO THE SPAM FILTERING PROBLEM

E-Mail spam is one of the major problems plaguing the contemporary Internet, causing an inconvenience to an individual user and financial loss to a company. Spam filtering allows for early detection of unwanted messages and separates them from the incoming e-mail. Nonetheless, designing an effective spam detection system is not a trivial task, due to the problems connected with the analysis of the e-mail content and the occurrence of variation in spam characteristics. This article presents an application of a novel ensemble classifier system for spam detection. The system is an extension of the adaptive splitting and selection (AdaSS) framework. The idea of the ensemble is based on the assumption that high effectiveness of detection can be obtained by exploitation of the local competency of a set of diverse elementary classifiers. Therefore, the ensemble training algorithm divides the feature space into several disjoint subspaces and assigns an area classifier to each of them. The area classifier consists of elementary classifiers that make a collective decision based on the weighted fusion of their support functions. The weight reflects the local competency of the classifier. To maintain the diversity of the pool of elementary classifiers, we exploit different e-mail feature extraction methods while filling the pool. There are two main extensions of the presented algorithm over original AdaSS: the aforementioned weighted fusion model used for decision making and adaptation of the AdaSS training procedure to process data streams featuring the concept drift. The effectiveness of the classifier model in spam recognition was verified in a series of experiments on two sets of spam databases. Comparison of the algorithm with some other state-of-the-art ensemble methods showed that the presented AdaSS extension can effectively recognize local competences of elementary classifiers and result in very high effectiveness of spam recognition outperforming competing methods.