Matej Mertik

Estimating Software Quality with Advanced Data Mining Techniques

Current software quality estimation models often involve the use of supervised learning methods for building a software fault prediction models. In such models, dependent variable usually represents a software quality measurement indicating the quality of a module by risk-basked class membership, or the number of faults. Independent variables include various software metrics as McCabe, Error Count, Halstead, Line of Code, etc... In this paper we present the use of advanced tool for data mining called Multimethod on the case of building software fault prediction model. Multimethod combines different aspects of supervised learning methods in dynamical environment and therefore can improve accuracy of generated prediction model. We demonstrate the use Multimethod tool on the real data from the Metrics Data Project Data (MDP) Repository. Our preliminary empirical results show promising potentials of this approach in predicting software quality in a software measurement and quality dataset.

Using Visual Interpretation of Small Ensembles in Microarray Analysis

Many different classification models and techniques have been employed on gene expression data. These computational methods are in rapid and continuous evolution and there is no clear consensus on which methods are best to cope with the complex microarray data analysis. Currently ensembles of classifiers are regarded as one of the best classification techniques as they can achieve excellent classification accuracy in comparison to single classifiers methods. One of their main drawbacks is their incomprehensibility. This paper addresses the important issue of the tradeoff between accuracy and comprehensibility when building ensembles and proposes a novel visual technique for interactive interpretation of the knowledge from the small ensembles consisting of only a few decision trees. This way we can achieve better accuracy compared to single classifier, but still maintain a certain level of comprehensibility in small ensembles. The results show that our small ensembles outperform the single classifiers and still retain comprehensibility. Our study also points out that in order to take advantage of our proposed method we need more effective small ensemble building techniques

Using LSTM recurrent neural networks for monitoring the LHC superconducting magnets

Abstract The superconducting LHC magnets are coupled with an electronic monitoring system which records and analyzes voltage time series reflecting their performance. A currently used system is based on a range of preprogrammed triggers which launches protection procedures when a misbehavior of the magnets is detected. All the procedures used in the protection equipment were designed and implemented according to known working scenarios of the system and are updated and monitored by human operators. This paper proposes a novel approach to monitoring and fault protection of the Large Hadron Collider (LHC) superconducting magnets which employs state-of-the-art Deep Learning algorithms. Consequently, the authors of the paper decided to examine the performance of LSTM recurrent neural networks for modeling of voltage time series of the magnets. In order to address this challenging task different network architectures and hyper-parameters were used to achieve the best possible performance of the solution. The regression results were measured in terms of RMSE for different number of future steps and history length taken into account for the prediction. The best result of RMSE = 0 . 00104 was obtained for a network of 128 LSTM cells within the internal layer and 16 steps history buffer.

iAPERAS - intelligent athlete's personal assistant

Non-professional athletes usually rely on the information about training methods and nutrition recommendations provided online. However, the quality of online information sources is extremely variable. iAPERAS is an expert system using Bayes networks and designed for athletes. It represents a better alternative to online resources, because it is based on scientific research findings and evaluated by domain experts.

The model of an anomaly detector for HiLumi LHC magnets based on Recurrent Neural Networks and adaptive quantization

This paper focuses on an examination of an applicability of Recurrent Neural Network models for detecting anomalous behavior of the CERN superconducting magnets. In order to conduct the experiments, the authors designed and implemented an adaptive signal quantization algorithm and a custom GRU-based detector and developed a method for the detector parameters selection. Three different datasets were used for testing the detector. Two artificially generated datasets were used to assess the raw performance of the system whereas the 231 MB dataset composed of the signals acquired from HiLumi magnets was intended for real-life experiments and model training. Several different setups of the developed anomaly detection system were evaluated and compared with state-of-the-art OC-SVM reference model operating on the same data. The OC-SVM model was equipped with a rich set of feature extractors accounting for a range of the input signal properties. It was determined in the course of the experiments that the detector, along with its supporting design methodology, reaches F1 equal or very close to 1 for almost all test sets. Due to the profile of the data, the best_length setup of the detector turned out to perform the best among all five tested configuration schemes of the detection system. The quantization parameters have the biggest impact on the overall performance of the detector with the best values of input/output grid equal to 16 and 8, respectively. The proposed solution of the detection significantly outperformed OC-SVM-based detector in most of the cases, with much more stable performance across all the datasets.Abstract This paper focuses on an examination of an applicability of Recurrent Neural Network models for detecting anomalous behavior of the CERN superconducting magnets. In order to conduct the experiments, the authors designed and implemented an adaptive signal quantization algorithm and a custom Gated Recurrent Unit-based detector and developed a method for the detector parameters selection. Three different datasets were used for testing the detector. Two artificially generated datasets were used to assess the raw performance of the system whereas the dataset intended for real-life experiments and model training was composed of the signals acquired from a new type of magnet, to be used during High-Luminosity Large Hadron Collider project. Several different setups of the developed anomaly detection system were evaluated and compared with state-of-the-art One Class Support Vector Machine (OC-SVM) reference model operating on the same data. The OC-SVM model was equipped with a rich set of feature extractors accounting for a range of the input signal properties. It was determined in the course of the experiments that the detector, along with its supporting design methodology, reaches F1 equal or very close to 1 for almost all test sets. Due to the profile of the data, the setup with the lowest maximum false anomaly length of the detector turned out to perform the best among all five tested configuration schemes of the detection system. The quantization parameters have the biggest impact on the overall performance of the detector with the best values of input/output grid equal to 16 and 8, respectively. The proposed solution of the detection significantly outperformed OC-SVM-based detector in most of the cases, with much more stable performance across all the datasets.

Using Cellular Automata for feature construction - preliminary study

When first faced with a learning task, it is often not clear what a good representation of the training data should look like. We are often forced to create some set of features that appear plausible, without any strong confidence that they will yield superior learning. Beside, we often do not have any prior knowledge of what learning method is the best to apply, and thus often try multiple methods in an attempt to find the one that performs best. This paper describes a new method and its preliminary study for constructing features based on cellular automata (CA). Our approach uses self-organisation ability of cellular automata by constructing features being most efficient for making predictions. We present and compare the CA approach with standard genetic algorithm (GA) which both use genetic programming (GP) for constructing the features. We show and discuss some interesting properties of using CA approach in our preliminary experimental study by constructing features on synthetically generated dataset and benchmark datasets from the UCI machine learning repository. Based on the interesting results, we conclude with directions and orientation of the future work with ideas of applicability of CA approach in the feature.

Improving Clinical Path Management Strategies by Constructive Induction

Digitalization of data has provided self understandable aspect of collect, store and retrieve large amounts of documents in databases, data repositories and data warehouses. Discovery of knowledge hidden in these databases can provide organizations with insight into there own internal intellectual assets. However, how to interpret meaningful information from this data remains one of the important challenges. This paper emphasizes importance of the preprocess part of KDD in such large data repositories. A health care organization study is used as an illustrative example where the KDD methods of C5.0 and CART are used. With careful analysis on large data, such as the use of proper feature sets, appropriate sampling sets, important meaningful information as insight into clinical pathways are discovered.

DETECTING FAULT MODULES USING BIOINFORMATICS TECHNIQUES

Many software reliability studies attempt to develop a model for predicting the faults of a software module because the application of good prediction models provides important information on significant metrics that should be observed in the early stages of implementation during software development. In this article we propose a new method inspired by a multi-agent based system that was initially used for classification and attribute selection in microarray analysis. Best classifying gene subset selection is a common problem in the field of bioinformatics. If we regard the software metrics measurement values of a software module as a genome of that module, and the real world dynamic characteristic of that module as its phenotype (i.e. failures as disease symptoms) we can borrow the established bioinformatics methods in the manner first to predict the module behavior and second to data mine the relations between metrics and failures.