Spiros Likothanassis

Integrating feature and instance selection for text classification

Instance selection and feature selection are two orthogonal methods for reducing the amount and complexity of data. Feature selection aims at the reduction of redundant features in a dataset whereas instance selection aims at the reduction of the number of instances. So far, these two methods have mostly been considered in isolation. In this paper, we present a new algorithm, which we call FIS (Feature and Instance Selection) that targets both problems simultaneously in the context of text classificationOur experiments on the Reuters and 20-Newsgroups datasets show that FIS considerably reduces both the number of features and the number of instances. The accuracy of a range of classifiers including Naïve Bayes, TAN and LB considerably improves when using the FIS preprocessed datasets, matching and exceeding that of Support Vector Machines, which is currently considered to be one of the best text classification methods. In all cases the results are much better compared to Mutual Information based feature selection. The training and classification speed of all classifiers is also greatly improved.

Scalable association-based text classification

Naive Bayes (NB) classifier has long been considered a core methodology in text classification mainly due to its simplicity and computational efficiency. There is an increasing need however for methods that can achieve higher classification accuracy while maintaining the ability to process large document collections. In this paper we examine text categorization methods from a perspective that considers the tradeoff between accuracy and scalability to large data sets and large feature sizes. We start from the observation that Support Vector Machines, one of the best text categorization methods cannot scale up to handle the large document collections involved in many real word problems. We then consider bayesian extensions to NB that achieve higher accuracy by relaxing its strong independence assumptions. Our experimental results show that LB, an association-based lazy classifier can achieve a good tradeoff between high classification accuracy and scalability to large document collections and large feature sizes.

YamiPred: A Novel Evolutionary Method for Predicting Pre-miRNAs and Selecting Relevant Features

MicroRNAs (miRNAs) are small non-coding RNAs, which play a significant role in gene regulation. Predicting miRNA genes is a challenging bioinformatics problem and existing experimental and computational methods fail to deal with it effectively. We developed YamiPred, an embedded classification method that combines the efficiency and robustness of support vector machines (SVM) with genetic algorithms (GA) for feature selection and parameters optimization. YamiPred was tested in a new and realistic human dataset and was compared with state-of-the-art computational intelligence approaches and the prevalent SVM-based tools for miRNA prediction. Experimental results indicate that YamiPred outperforms existing approaches in terms of accuracy and of geometric mean of sensitivity and specificity. The embedded feature selection component selects a compact feature subset that contributes to the performance optimization. Further experimentation with this minimal feature subset has achieved very high classification performance and revealed the minimum number of samples required for developing a robust predictor. YamiPred also confirmed the important role of commonly used features such as entropy and enthalpy, and uncovered the significance of newly introduced features, such as %A-U aggregate nucleotide frequency and positional entropy. The best model trained on human data has successfully predicted pre-miRNAs to other organisms including the category of viruses.

A Hybrid Support Vector Fuzzy Inference System for the Classification of Leakage Current Waveforms Portraying Discharges

Abstract Several techniques have been applied on leakage current waveforms in order to extract information regarding electrical activity on high-voltage insulators. However, a fully representative value is yet to be defined. In this article, a hybrid support vector fuzzy inference system is introduced as a classification tool. The system incorporates fuzzy logic, genetic algorithms, and support vector machines. Apart from the classification accuracy achieved, the system also produces a set of fuzzy rules under which the classification is made, allowing a further insight of the process. A comparison is made to other classification tools previously applied on the same data set.

Application service provision through the grid: business models and architectures

The new economy is the result of the information revolution that promotes the emergence of networked, ubiquitous and collaborative service provision. Furthermore, the globalisation of trade has increased the number of competitors, driving the need for federated and networked business models. In this paper we argue that next generation ASP models could benefit from the use of the grid. We describe a grid-enabled architecture for coordinated resource sharing and problem solving in dynamic, multi-institutional ASP vendors.

Predicting protein complexes from weighted protein-protein interaction graphs with a novel unsupervised methodology

OBJECTIVE Proteins are considered to be the most important individual components of biological systems and they combine to form physical protein complexes which are responsible for certain molecular functions. Despite the large availability of protein-protein interaction (PPI) information, not much information is available about protein complexes. Experimental methods are limited in terms of time, efficiency, cost and performance constraints. Existing computational methods have provided encouraging preliminary results, but they phase certain disadvantages as they require parameter tuning, some of them cannot handle weighted PPI data and others do not allow a protein to participate in more than one protein complex. In the present paper, we propose a new fully unsupervised methodology for predicting protein complexes from weighted PPI graphs. METHODS AND MATERIALS The proposed methodology is called evolutionary enhanced Markov clustering (EE-MC) and it is a hybrid combination of an adaptive evolutionary algorithm and a state-of-the-art clustering algorithm named enhanced Markov clustering. EE-MC was compared with state-of-the-art methodologies when applied to datasets from the human and the yeast Saccharomyces cerevisiae organisms. RESULTS Using public available datasets, EE-MC outperformed existing methodologies (in some datasets the separation metric was increased by 10-20%). Moreover, when applied to new human datasets its performance was encouraging in the prediction of protein complexes which consist of proteins with high functional similarity. In specific, 5737 protein complexes were predicted and 72.58% of them are enriched for at least one gene ontology (GO) function term. CONCLUSIONS EE-MC is by design able to overcome intrinsic limitations of existing methodologies such as their inability to handle weighted PPI networks, their constraint to assign every protein in exactly one cluster and the difficulties they face concerning the parameter tuning. This fact was experimentally validated and moreover, new potentially true human protein complexes were suggested as candidates for further validation using experimental techniques.

Modeling and Trading FTSE100 Index Using a Novel Sliding Window Approach Which Combines Adaptive Differential Evolution and Support Vector Regression

The motivation for this paper is to introduce a novel short term trading strategy using a machine learning based methodology to model the FTSE100 index. The proposed trading strategy deploys a sliding window approach to modeling using a combination of Differential Evolution and Support Vector Regressions. These models are tasked with forecasting and trading daily movements of the FTSE100 index. To test the efficiency of our proposed method, it is benchmarked against two simple trading strategies (Buy and Hold and Naive Strategy) and two modern machine learning methods. The experimental results indicate that the proposed method outperformsall other examined models in terms of statistical accuracy and profitability. As a result, this hybrid approach is established as a credible and worth trading strategy when applied to time series analysis.

ncRNAclass: A Web Platform for Non-Coding RNA Feature Calculation and MicroRNAs and Targets Prediction

According to the central dogma of Biology it was commonly accepted that most of the genetic information was transacted by proteins. Recent experimental techniques revealed that the majority of mammalian genomes and other complex organisms are in fact transcribed into non-coding RNAs. Typically, non-coding RNAs are small nucleotide sequences that are not transcribed into proteins and have a profound regulatory role. Present advances in computational biosciences linked their abnormal functionality to many diseases and re-stated the principles of basic therapeutic strategies. The effective identification of non-coding RNAs and their biological role emerges as a new and challenging bioinformatics problem. ncRNAclass (http://biotools.ceid.upatras.gr/ncrnaclass/) is a web platform that allows for efficient computation of a set of features that can describe effectively the broad class of non-coding RNAs. Moreover, it enables the calculation of features that include information about the targeting behavior of miRNAs. The tool operates under a user-friendly interface and its pilot implementation incorporates prediction models for the well-known class of microRNAs and for prediction their mRNA targets. The prediction models are based on two novel evolutionary Machine Learning algorithms that achieve very high classification performance in comparison with existing methods. The platform is also equipped with a data warehouse, with manually curated sequences, that enables fast information retrieval and data mining utilities.

The Human Interactome Knowledge Base (HINT-KB): an integrative human protein interaction database enriched with predicted protein---protein interaction scores using a novel hybrid technique

Proteins are the functional components of many cellular processes and the identification of their physical protein–protein interactions (PPIs) is an area of mature academic research. Various databases have been developed containing information about experimentally and computationally detected human PPIs as well as their corresponding annotation data. However, these databases contain many false positive interactions, are partial and only a few of them incorporate data from various sources. To overcome these limitations, we have developed HINT-KB (http://biotools.ceid.upatras.gr/hint-kb/), a knowledge base that integrates data from various sources, provides a user-friendly interface for their retrieval, calculates a set of features of interest and computes a confidence score for every candidate protein interaction. This confidence score is essential for filtering the false positive interactions which are present in existing databases, predicting new protein interactions and measuring the frequency of each true protein interaction. For this reason, a novel machine learning hybrid methodology, called (Evolutionary Kalman Mathematical Modelling—EvoKalMaModel), was used to achieve an accurate and interpretable scoring methodology. The experimental results indicated that the proposed scoring scheme outperforms existing computational methods for the prediction of PPIs.

Modelling and Trading the DJIA Financial Index Using Neural Networks Optimized with Adaptive Evolutionary Algorithms

In the current paper we study an evolutionary framework for the optimization of various types Neural Networks’ structure and parameters. Two different adaptive evolutionary algorithms, named as adaptive Genetic Algorithms (aGA) and adaptive Differential Evolution (aDE), were developed to optimize the structure and the parameters of two different types of Neural Networks: Multilayer Perceptron (MLPs) and Wavelet Neural Networks (WNN). Wavelet neural networks have been introduced as an alternative to MLPs to overcome their shortcomings presenting more compact architecture and higher learning speed. Furthermore, the evolutionary algorithms, which were implemented, are both adaptive in terms that their most important parameters (Mutation and Crossover probabilities) are assigned using a self adaptive scheme. The motivation of this paper is to uncover novel hybrid methodologies for the task of forecasting and trading DJIE financial index. This is done by benchmarking the forecasting performance the four proposed hybrid methodologies (aGA-MLP, aGA-WNN, aDE-MLP and aDE-WNN) with some traditional techniques, either statistical such as a an autoregressive moving average model (ARMA), or technical such as a moving average covcergence/divergence model (MACD). The trading performance of all models is investigated in a forecast and trading simulation on our time series over the period 1997-2012. As it turns out, the aDE-WNN hybrid methodology does remarkably well and outperforms all other models in simple trading simulation exercises. (This paper is submitted for the ACIFF workshop).