Ensemble Learning

Motivation: Three major problems confront the construction of a human genetic network from heterogeneous genomics data using kernel-based approaches: definition of a robust gold-standard negative set, large-scale learning and massive missing data values. Results: The proposed graph-based approach generates a robust GSN for the training process of genetic network construction. The RVM-based ensemble model that combines AdaBoost and reduced-feature yields improved performance on large-scale learning problems with massive missing values in comparison to Naïve Bayes.

Hedonic pricing models attempt to model a relationship between object attributes and the object's price. Traditional hedonic pricing models are often parametric models that suffer from misspecification. In this paper we create these models by means of boosted CART models. The method is explained in detail and applied to various datasets. Empirically, we find substantial reduction of errors on out-of-sample data for two out of three datasets compared with a stepwise linear regression model. We interpret the boosted models by partial dependence plots and relative importance plots. This reveals some interesting nonlinearities and differences in attribute importance across the model types. Categorical Name Levels Country France (13) Germany (40) Italy (3) Japan (99) USA (20) UK (3) Sweden (17) FuelType diesel (20) gas (185) Aspiration standard (168) turbo (37) BodyStyle convertible (6) hardtop (8) hatchback (70) sedan (96) wagon Drive-wheels 4wd (9) fwd (120) rwd (76) EngineLocation front (202) rear EngineType dohc (12) dohcv (1) l (12) ohc (148) ohcf (15) ohcv (13) rotor (4) FuelSystem 1bbl 2bbl 4bbl idi mfi (1) mpfi (94) spdi (9) spfi (1) Numerical Name Avg Stdv Min Max (Missing)

Forests are of great importance for the sustainability of the ecosystem as well as for the mankind. The poplar species in the forest ecosystems are one of the most valuable and beneficial species for the society and environment. Turkey is a very rich country in terms of cultivated poplar species. The determination of poplar areas in Turkey is usually based on field studies. Although these methods require high cost, time and labor need, the results obtained vary and are insufficient in terms of accuracy. Determination of poplar cultivated areas and mapping of their spatial locations play an important role for decision-makers and planners to enhance the economic and ecological value of poplar trees. The main goal of this study is to map Poplar (P.deltoides) cultivated areas in Akyazi district of Sakarya, Turkey province using the Sentinel-2 satellite imagery. For this purpose, object-based classification based on multi-resolution segmentation algorithm was utilized to create image objects and then, two outstanding ensemble learning algorithms, Random Forest (RF) and Rotation Forest (RotFor) were applied to produce thematic maps. In order to analyze the effects of the spectral bands of the Sentinel-2 image on the object-based classification performance, two datasets consisting of different spectral band combinations (i.e. four 10m bands and 10m ten pan-sharpened bands) were used. The performance evaluation results showed that the RotFor classifier produced superior classification performances compared to the RF classifier for the band combinations considered in this study. It was also observed that the difference in classification accuracy reached to approximately 6% in terms of overall accuracy. Moreover, results that Sentinel-2 image dataset having 4-band dataset and 10-band dataset was effective for determination of poplar areas and the class level accuracy reached to ~99% in terms of F-score.

Evolutionary Learning proceeds by evolving a population of classifiers, from which it generally returns (with some notable exceptions) the single best-of-run classifier as final result. In the meanwhile, Ensemble Learning, one of the most efficient approaches in supervised Machine Learning for the last decade, proceeds by building a population of diverse classifiers. Ensemble Learning with Evolutionary Computation thus receives increasing attention. The Evolutionary Ensemble Learning (EEL) approach presented in this paper features two contributions. First, a new fitness function, inspired by co-evolution and enforcing the classifier diversity, is presented. Further, a new selection criterion based on the classification margin is proposed. This criterion is used to extract the classifier ensemble from the final population only (Off-EEL) or incrementally along evolution (On-EEL). Experiments on a set of benchmark problems show that Off-EEL outperforms single-hypothesis evolutionary learning and state-of-art Boosting and generates smaller classifier ensembles.

Recruitment in the IT sector has been on the rise in recent times. Software companies are on the hunt to recruit raw talent right from the colleges through job fairs. The process of allotment of projects to the new recruits is a manual affair, usually carried out by the Human Resources department of the organization. This process of project allotment to the new recruits is a costly affair for the organization as it relies mostly on human effort. In the recent times, software companies round the world are leveraging the advances in machine learning and Artificial intelligence in general to automate routine tasks in the enterprise to increase the productivity. In the paper, we discuss the design and implementation of a resume classifier application which employs an ensemble learning based voting classifier to classify a profile of a candidate into a suitable domain based on his interest, work-experience and expertise mentioned by the candidate in the profile. The model employs topic modelling techniques to introduce a new domain to the list of domains upon failing to achieve the threshold value of confidence for the classification of the candidate profile. The Stack-Overflow REST APIs are called for the profiles which fail on the confidence threshold test set in the application. The topics returned by the APIs are subjected to topic modelling to obtain a new domain, on which the voting classifier is retrained after a fixed interval to improve the accuracy of the model.Overall, emphasis is laid out on building a dynamic machine learning automation tool which is not solely dependent on the training data in allotment of projects to the new recruits. We extended our previous work withnew learning model that has the ability to classify the resumes with better accuracy and support more new domains.

Company bankruptcy is often a very big problem for companies. The impact of bankruptcy can cause losses to elements of the company such as owners, investors, employees, and consumers. One way to prevent bankruptcy is to predict the possibility of bankruptcy based on the company's financial data. Therefore, this study aims to find the best predictive model or method to predict company bankruptcy using the dataset from Polish companies bankruptcy. The prediction analysis process uses the best feature selection and ensemble learning. The best feature selection is selected using feature importance to XGBoost with a weight value filter of 10. The ensemble learning method used is stacking. Stacking is composed of the base model and meta learner. The base model consists of K-nearest neighbor, decision tree, support vector machines (SVM), and random forest, while the meta learner used is LightGBM. The stacking model accuracy results can outperform the base model accuracy with an accuracy rate of 97%.

In this study, an empirical mode decomposition (EMD) based neural network ensemble learning paradigm is proposed for world crude oil spot price forecasting. For this purpose, the original crude oil spot price series were first decomposed into a finite, and often small, number of intrinsic mode functions (IMFs). Then a three-layer feedforward neural network (FNN) model was used to model each of the extracted IMFs, so that the tendencies of these IMFs could be accurately predicted. Finally, the prediction results of all IMFs are combined with an adaptive linear neural network (ALNN), to formulate an ensemble output for the original crude oil price series. For verification and testing, two main crude oil price series, West Texas Intermediate (WTI) crude oil spot price and Brent crude oil spot price, are used to test the effectiveness of the proposed EMD-based neural network ensemble learning methodology. Empirical results obtained demonstrate attractiveness of the proposed EMD-based neural network ensemble learning paradigm.

Nitrogen (N) has been linked to different ecosystem processes, and retrieving this important foliar biochemical constituent from remote sensing observations is of widespread interest. Since N is not explicitly represented in physically based radiative transfer models, empirical methods have been used as an alternative. The spectral bands selected during the calibration of empirical methods have been interpreted in the context of light-N interactions and, consequently, ecosystem function. The low amount of leaves on shrubs and their sparse distribution in drylands create an environment, in which the canopy structure and the bright background soil play an important role in the canopy total radiation budget. In this paper, we examine the assumption that removing the impact of canopy structure and soil will result in improved N retrieval using the empirical methods. We report the inconsistencies in the selection of spectral bands among the empirical approaches. Moreover, these methods are sensitive to the scale of the study and band transformations. Using the generalized theory of canopy spectral invariants, we found that at the canopy scale, a combination of canopy structure and soil dominates the total canopy radiation. At the plot scale, soil

Random Forest is an ensemble machine learning method developed by Leo Breiman in 2001. Since then, it has been considered the state-of-the-art solution in machine learning applications. Compared to the other ensemble methods, random forests exhibit superior predictive performance. However, empirical and statistical studies prove that the random forest algorithm generates unnecessarily large number of base decision trees. This may cost high computational efficiency, predictive time, and occasional decrease in effectiveness. In this paper, Authors survey existing random forest pruning techniques and compare the performance between them. The research revolves around both the static and dynamic pruning technique and analyses the scope of improving the performance of random forest by techniques including generating diverse and accurate decision trees, selecting high performance subset of decision trees, genetic algorithms and other state of art methods, among others.

Ensemble learning strategies, especially Boosting and Bagging decision trees, have demonstrated impressive capacities to improve the prediction accuracy of base learning algorithms. Further gains have been demonstrated by strategies that combine simple ensemble formation approaches. In this paper, we investigate the hypothesis that the improvement in accuracy of multistrategy approaches to ensemble learning is due to an increase in the diversity of ensemble members that are formed. In addition, guided by this hypothesis, we develop three new multistrategy ensemble learning techniques. Experimental results in a wide variety of natural domains suggest that these multistrategy ensemble learning techniques are, on average, more accurate than their component ensemble learning techniques.

Ensemble learning is one of machine learning method that can solve performance measurement problem. Standalone classifiers often show a poor performance result, thus why combining them with ensemble methods can improve their performance scores. Ensemble learning has several methods, in this study, three methods of ensemble learning are compared with standalone classifiers of support vector machine, Naïve Bayes, and decision tree. bagging, AdaBoost, and voting are the ensemble methods that are combined then compared to standalone classifiers. From 1670 dataset of twitter mentions about tourist's attraction, ensemble methods did not show a specific improvement in accuracy and precision measurement since it generated the same result as decision tree as standalone classifier. Bagging method showed a significant development in recall, f-measure, and area under curve (AUC) measurement. For overall performance, decision tree as standalone classifier and decision tree with AdaBoost method have the highest score for accuracy and precision measurements, meanwhile support vector machine with bagging method has the highest score for recall, f-measure, and AUC.

—Early detection of cancer is the most promising way to enhance a patient's chance for survival. This paper presents a computer-aided classification method using computed tomography (CT) images of the lung based on ensemble of three classifiers including MLP, KNN and SVM. In this study, the entire lung is first segmented from the CT images and specific features like Roundness, Circularity, Compactness, Ellipticity, and Eccentricity are calculated from the segmented images. These morphological features are used for classification process in a way that each classifier makes its own decision. Finally, majority voting method is used to combine decisions of this ensemble system. The performance of this system is evaluated using 60 CT scans collected by Lung Image Database Consortium (LIDC) and the results show good improvement in diagnosing of pulmonary nodules.

This paper is prepared to provide a brief introduction to the topic of Ensemble Learning. It aims to provide the reader with a broad overview on the approach of Ensemble Methods.

Sections:

-What is Ensemble Learning?

-The Rationale Behind Ensemble Methods

-Common Approaches To Ensemble Methods

-Success Factors Of Ensemble Methods

-Proven Applications of Ensemble Methods

Note: This paper is part of EVOLUTIQ's educational series on the subject of systematic trading. These papers will discuss different approaches to systematic trading as well as present specific trading strategies and associated risk management techniques.

Context: Several issues hinder software defect data including redundancy, correlation, feature irrelevance and missing samples. It is also hard to ensure balanced distribution between data pertaining to defective and non-defective software. In most experimental cases, data related to the latter software class is dominantly present in the dataset. Objective: The objectives of this paper are to demonstrate the positive effects of combining feature selection and ensemble learning on the performance of defect classification. Along with efficient feature selection, a new two-variant (with and without feature selection) ensemble learning algorithm is proposed to provide robustness to both data imbalance and feature redundancy. Method: We carefully combine selected ensemble learning models with efficient feature selection to address these issues and mitigate their effects on the defect classification performance. Results: Forward selection showed that only few features contribute to high area under the receiveroperating curve (AUC). On the tested datasets, greedy forward selection (GFS) method outperformed other feature selection techniques such as Pearson's correlation. This suggests that features are highly unstable. However, ensemble learners like random forests and the proposed algorithm, average probability ensemble (APE), are not as affected by poor features as in the case of weighted support vector machines (W-SVMs). Moreover, the APE model combined with greedy forward selection (enhanced APE) achieved AUC values of approximately 1.0 for the NASA datasets: PC2, PC4, and MC1. Conclusion: This paper shows that features of a software dataset must be carefully selected for accurate classification of defective components. Furthermore, tackling the software data issues, mentioned above, with the proposed combined learning model resulted in remarkable classification performance paving the way for successful quality control.

The objective of the present thesis is the design, development and evaluation of a breast
cancer detection model through applying machine learning techniques on data from blood
analysis and anthropometric data.
For the development and evaluation of the model the Coimbra Breast Cancer Dataset
has been used, which is publicly available through the UCI Machine Learning Repository.
This dataset consists of 116 instances. Out of the 116 women participants, 64 had been
diagnosed with breast cancer . For each participant the values of serum Glucose, Leptin,
Adiponectin, Resistin and Chemokine Monocyte Chemoattractant Protein 1 aMCP-1), and
levels of plasma Insulin were assessed by blood analysis. The Homeostasis Model
Assessment aHOMA) index and the Leptin/Antiponectin ratios were calculated. Moreover,
anthropometric data such as age and Body Mass Index aBMI) were recorded for each
participant.
For the development of the model, the use of several machine learning models
aSVM, Logistic Regression, Random Forest, XGBoost, Naive Bayes) has been investigated,
while extended search of the optimal combination of biomarkers for each model has been
conducted. Aiming at achieving high levels of accuracy, ensemble learning techniques have
been applied in order to combine the outputs of the primary models towards producing the
final decision. The model which achieved the highest accuracy was based on the combined
use of the XGBoost, Random Forest, Support Vector Machines and KNN models and the
application of the Dynamic Weighted Voting based on Accuracy on the outputs of the primary
models. The obtained results demonstrate the effectiveness of the model aAUC : 0.928 ,
Accuracy : 92.24% , Sensitivity : 93.75%, Specificity 90.38%, F1-score 0.930 and Precision
0.923).

In this paper, a relevance vector machine based infinite decision agent ensemble learning (RVM Ideal) system is proposed for the robust credit risk analysis. In the first level of our model, we adopt soft margin boosting to overcome overfitting. In the second level, the RVM algorithm is revised for boosting so that different RVM agents can be generated from the updated instance space of the data. In the third level, the perceptron Kernel is employed in RVM to simulate infinite subagents. Our system RVM Ideal also shares some good properties, such as good generalization performance, immunity to overfitting and predicting the distance to default. According to the experimental results, our proposed system can achieve better performance in term of sensitivity, specificity and overall accuracy.

In this era of big data, classifying imbalanced real-life data in supervised learning is a challenging research issue. Standard data sampling methods: under-sampling and over-sampling have several limitations for dealing with big data. Mostly, under-sampling approach removes data points from majority class instances and over-sampling approach engenders artificial minority class instances to make the data balanced. However, we may lose informative information instances using under-sampling approach, and under other conditions over-sampling approach causes overfitting problem. In this research work, we have presented a new cluster-based under-sampling approach by amalgamating ensemble learning (e.g. RandomForest classifier) for classification of imbalanced data that we implemented in Julia. We have collected actual illegal money transaction telecom fraud data, which is highly imbalanced with only 8,213 minority class instances among 63,62,620 instances. The proposed method bifurcates the data into majority class and minority class instances. Then, clusters the majority class instances into several clusters and considers a set of instances from each cluster to create several sub-balanced datasets. Finally, a number of classifiers are generated using these balances datasets and apply majority voting technique for classifying unknown new instances. We have tested the proposed method on a separate test dataset that achieved 95.82% accuracy.

Ensemble  methods are  popular  strategies for improving the  predictive ability  of a machine learning  model.  An ensemble consists  of  a  set of  individually  trained base learners/models whose  predictions  are  combined  when  classifying  new  cases. Previous researches have  shown that an ensemble is  on the  average  more  accurate than a  single base  model. Bagging, Boosting and  Stacking  are  some  popular  ensemble  techniques which we  studied  in this paper. We  evaluated  these  ensembles  on 9  data  sets. From our results, we  observed the following. First,  an ensemble is always more  accurate than a single  base  model.  Secondly, we  observed that Boosting ensembles is  on the average better than Bagging while  Stacking  (meta-learning)  is on  the average  more  accurate than Boosting and  Bagging. Further experiment also shows that  the gain in predictive  power of  any  ensembles may  sometimes be small  or even  decrease  depending on the  data set.

In the cyber era, Machine Learning (ML) has provided us with the solutions to these problems with the implementation of Gradient Boosting Machines (GBM). We have ample algorithms to choose from to do gradient boosting for our training data but still, we encounter different issues like poor accuracy, high loss, large variance in the result. Here, we are going to introduce you to a state of the art machine learning algorithm XGBoost built by Tianqi Chen, that will not only overcome the issues but also perform exceptionally well for regression and classification problems. This blog will help you discover the insights, techniques, and skills with XGBoost that you can then bring to your machine learning projects.
eXtreme Gradient Boosting (XGBoost) is a scalable and improved version of the gradient boosting algorithm (terminology alert) designed for efficacy, computational speed and model performance. It is an open-source library and a part of the Distributed Machine Learning Community. XGBoost is a perfect blend of software and hardware capabilities designed to enhance existing boosting techniques with accuracy in the shortest amount of time. Here’s a quick look at an objective benchmark comparison of XGBoost with other gradient boosting algorithms trained on random forest with 500 trees, performed by Szilard Pafka.

In this study the ability of Ensemble Deep Learning Network in forecasting the daily closing price of Nigerian Stock Exchange was investigated. An ensemble method consisting of several deep learning network which are feed forward neural network (FFN), recurrent neural network (RNN) and cascade forward network (CFN) were used to train the outputs of Levenberg-Marquardt (LM), Resilient Back Propagation (RBP) and Scaled Conjugate Gradient (SCG)for enhanced performance. An ensemble deep-supervised was proposed by combining deep learning and supervised learning to achieve an optimal result. The dataset used was obtained from historical closing stock prices with integrating five technical analysis as inputs. Daily stock exchange rates of first city monument bank and zenith bank extracted from Nigerian Stock are used for testing the model prediction ability. Experimental result analysis shows that our proposed model has 99% accuracy higher than the three deep learning network used in this study.

The discovery of a bizarre manuscript arrangement of the spurious Prelude and Fugue in B-flat major on the name BACH (BWV 898) for 32 hands (8 pianos, 16 players) led to a series of questions: What would have been the purpose of such an arrangement? Why was it written for what seems to be an extraordinarily impractical number of performers and instruments? Who was the arranger? Was it ever performed? If so, who were the performers and where did all the pianos come from? And why was this particular work chosen? Does this particular source give an additional, and previously unheeded, dimension to Bohemian Bach reception, and to Bach reception in general? These are just some of the enigmas posed by this peculiar find, which are explored in this article.

Satellite time-series data are bolstering global change research, but their use to elucidate land changes and vegetation dynamics is sensitive to algorithmic choices. Different algorithms often give inconsistent or sometimes conflicting interpretations of the same data. This lack of consensus has adverse implications and can be mitigated via ensemble modeling, an algorithmic paradigm that combines many competing models rather than choosing only a single "best" model. Here we report one such time-series decomposition algorithm for deriving nonlinear ecosystem dynamics across multiple timescales-A Bayesian Estimator of Abrupt change, Seasonal change, and Trend (BEAST). As an ensemble algorithm, BEAST quantifies the relative usefulness of individual decomposition models, leveraging all the models via Bayesian model averaging. We tested it upon simulated, Landsat, and MODIS data. BEAST detected changepoints, seasonality, and trends in the data reliably; it derived realistic nonlinear trends and credible uncertainty measures (e.g., occurrence probability of changepoints over time)-some information difficult to derive by conventional single-best-model algorithms but critical for interpretation of ecosystem dynamics and detection of low-magnitude disturbances. The combination of many models enabled BEAST to alleviate model misspecification, address algorithmic uncertainty, and reduce over-fitting. BEAST is generically applicable to time-series data of all kinds. It offers a new analytical option for robust changepoint detection and nonlinear trend analysis and will help exploit environmental time-series data for probing patterns and drivers of ecosystem dynamics.

Stock-market prediction using machine-learning technique aims at developing effective and efficient models that can provide a better and higher rate of prediction
accuracy. Numerous ensemble regressors and classifiers have been applied in stock
market predictions, using different combination techniques. However, three precarious issues come in mind when constructing ensemble classifiers and regressors. The first concerns with the choice of base regressor or classifier technique adopted. The second concerns the combination techniques used to assemble multiple regressors or classifiers and the third concerns with the quantum of regressors or classifiers to be ensembled. Subsequently, the number of relevant studies scrutinising these previously mentioned concerns are limited. In this study, we performed an extensive comparative analysis of ensemble techniques such as boosting, bagging, blending and super learners (stacking). Using Decision Trees (DT), Support Vector Machine (SVM) and Neural
Network (NN), we constructed twenty-five (25) different ensembled regressors and classifiers. We compared their execution times, accuracy, and error metrics over stock data from Ghana Stock Exchange (GSE), Johannesburg Stock Exchange (JSE), Bombay
Stock Exchange (BSE-SENSEX) and New York Stock Exchange (NYSE), from January 2012
to December 2018. The study outcome shows that stacking and blending ensemble techniques offer higher prediction accuracies (90–100%) and (85.7–100%) respectively, compared with that of bagging (53–97.78%) and boosting (52.7–96.32%). Furthermore, the root means square error (RMSE) recorded by stacking (0.0001–0.001) and blending (0.002–0.01) shows a better ft of ensemble classifiers and regressors based on these two techniques in market analyses compared with bagging (0.01–0.11) and boosting (0.01–0.443). Finally, the results undoubtedly suggest that an innovative study in the domain of stock market direction prediction ought to include ensemble techniques in their sets of algorithms.

In this paper, we demonstrate the use of a
multiple classifier system for classification of electroencephalogram
(EEG) signals. The main purpose of this
paper is to apply several approaches to classify motor
imageries originating from the brain in a more robust
manner. For this study, dataset II from BCI competition III
was used. To extract features from the brain signal, discrete
wavelet transform decomposition was used. Then, several
classic classifiers were implemented to be utilized in the
multiple classifier system, which outperforms the reported
results of other proposed methods on the dataset. Also, a
variety of classifier combination methods along with
genetic algorithm feature selection were evaluated and
compared in order to diminish classification error. Our
results suggest that an ensemble system can be employed to
boost EEG classification accuracy.

Badania prowadzone w ramach rozprawy doktorskiej dotyczyły bardzo istotnej, ze społecznego punktu widzenia, tematyki - zapobiegania chorobom układu krążenia, które są najczęstszą przyczyną zgonów na świecie.
Celem prowadzonych badań było stworzenie algorytmów służących do skutecznego, automatycznego rozpoznawania dysfunkcji mięśnia sercowego (prawidłowego rytmu zatokowego, rytmu z rozrusznika oraz 15 zaburzeń pracy serca). Algorytmy były projektowane z myślą o ich zastosowaniu w telemedycynie i urządzeniach mobilnych, wspomagających profilaktykę leczenia chorób układu krążenia. Główny nacisk został położony na skuteczność projektowanego rozwiązania, jednak ze względu na zamiar wykorzystania zaprojektowanych algorytmów w urządzaniach mobilnych dodatkowym kryterium oceny był czas trwania analizy która powinna odbywać się w czasie rzeczywistym. Należy nadmienić, że obecnie opisane, w literaturze naukowej, rozwiązania nie wykazują zadowalającej skuteczności.
Do zaprojektowania i przebadania algorytmów została stworzona baza danych zawierająca 744 fragmenty (10 sekundowe) sygnałów EKG od 29 pacjentów pochodzące z ogólnodostępnej bazy: MIH-BIH Arrhythmia Database. Analizowane fragmenty sygnałów EKG pochodziły tylko z jednego odprowadzenia: MLII.
W badaniach zastosowano oryginalną metodologię polegającą na analizie dłuższych (10 sekundowych) fragmentów sygnału EKG. W celu wzmocnienia charakterystycznych cech sygnału EKG estymowano widmową gęstość mocy za pomocą metody Welch'a i dyskretnej transformacji Fouriera. 
W ramach badań przeprowadzono 6 eksperymentów obejmujących podstawową analizę sygnałów EKG, genetyczną optymalizację parametrów i genetyczną selekcję cech oraz różne typy oryginalnych zespołów klasyfikatorów. Na podstawie przeprowadzonych eksperymentów przeanalizowano i wskazano najlepsze metody dotyczące: przetwarzania wstępnego sygnału, normalizacji, ekstrakcji i selekcji cech (składowych częstotliwościowych), walidacji krzyżowej, klasyfikatorów (SVM, kNN, PNN, RBFNN) i zespołów klasyfikatorów oraz optymalizacji parametrów. W wyniku szczegółowych badań została wytyczona ścieżka (kombinacja metod) dla której uzyskano najwyższą czułość rozpoznawania zaburzeń pracy mięśnia sercowego. Wyniki zweryfikowano za pomocą wyznaczonych współczynników (sumy błędów, czułości, współczynnika kappa, czasu klasyfikacji, itp.) uzyskanych na podstawie macierzy pomyłek.
Otrzymane wyniki w pełni potwierdzają słuszność prowadzonych badań i stawianych hipotez oraz udowadniają zrealizowanie celu - skutecznego i szybkiego rozpoznawania patologii pracy mięśnia sercowego. Najlepszy system - głęboki genetyczny zespół klasyfikatorów, uzyskał czułość rozpoznawania, aż 17 klas, na poziomie: SEN = 94.62% (ERRsum = 40 błędów na 744 klasyfikacje, skuteczność: ACC = 99.37%, specyficzność: SPE = 99.66%), przy czasie klasyfikowania jednego fragmentu sygnału EKG na poziomie: Ck = 0.8736 [s]. Na tle aktualnej literatury naukowej jest to bardzo dobry wynik (drugi najlepszy), potwierdzający sens prowadzenia dalszych badań w tej dziedzinie na podstawie zaproponowanej metodologii.

The growing dependence of modern society on telecommunication and information networks has become inevitable. Therefore, the security aspects of such networks play a strategic role in ensuring protection of data against misuse. Intrusion Detection systems (IDS) are meant to detect intruders who elude the "first line" protection. Data mining techniques are being used for building effective IDS. In this paper we analyze the performance of some data classifiers in a heterogeneous environment using voting ensemble system with the purpose of detecting anomaly based network intrusions. Experimental results using KDDCup 1999 benchmark dataset demonstrate that the voting ensemble technique yield significantly better results in detecting intrusions when compared to other techniques.

This paper tackles the problem of integrating household energy prosumers in Smart Energy Grids by analyzing a set of state-of-the-art energy forecasting techniques that allow individual or aggregated prosumers to evaluate their future energy demand and inform the Distributed System Operator (DSO) about potential grid imbalances. Thus, the DSO can perform a proactive strategy to manage the grid and avoid problems before they appear. The key element of this approach is the prediction technique, that must be accurate enough such that the resulting grid imbalances can be compensated in real-time. The paper evaluates a set of state-of-the-art statistical and Machine Learning (ML) prediction techniques, such as SARIMA, feed-forward and recurrent neural networks, support vector regression or ensemble prediction models, on real household historical energy demand logs by performing a feature selection process for each ML algorithm as to identify the best elements that influence the energy demand of a house. A set of experiments are performed on the REFIT Electrical Load Measurements data set evaluating each model's performance with respect to the selected features. Among the evaluated algorithms, the Ensemble Prediction Model gives best prediction accuracy, showing a Mean Absolute Percentage Error (MAPE) of 14.4% followed by the SVM model with a MAPE of 15.4%.

Improving accuracy, robustness and understandability is the objective of classification modeling. Regarding instability and performance limitation of existing rule learning techniques, we introduce an ensemble classifier based on randomized neighborhood reduction and neighborhood covering reduction. A set of reducts are obtained with randomized attribute reduction. A collection of rule sets are derived from the reducts based on neighborhood covering reduction. And then the classification result is output by combining the classification decision of different rule sets. The experiment result shows that the proposed technique is better than or equal to other classifiers, and is more stable when deals with no isy data.

Amaç: Günümüzde makine öğrenmesi yöntemleri hastalık tanısının konulmasında yaygın olarak kullanılmaktadır. Ancak sağlık verisinin büyük hacimli, çok boyutlu ve karmaşık olması nedeniyle dengesiz sınıf problemi ile karşılaşılması durumunda bu yöntemlerin doğrudan kulla-nımı performans düşüşüne neden olmaktadır. Bu çalışmada diyabet hastalarına ilişkin dengesiz yapıdaki bir veri seti kullanılarak çeşitli yeniden örnekleme yöntemleri dengesizlik problemi-nin giderilmesinde kullanılmış ve kolektif (ensemble) öğrenme algoritmalarına entegre edile-rek diyabet tanısı üzerinden sınıflandırma performansları karşılaştırılmıştır. Gereç Yöntemler: Kullanılan veriler Haziran-Eylül 2013 tarihleri arasında, İzmir Bozkaya Eğitim ve Araştırma Hastanesi, Endokrinoloji ve Metabolizma Hastalıkları polikliniğine başvuran, 18 yaşından büyük 185 hastadan elde edilmiştir. Diyabet tanısının sınıflandırmasına yönelik sınıf dengesizliği prob-lemini gidermek amacıyla alt örnekleme (under sampling), aşırı örnekleme (over sampling) ve sentetik azınlık aşırı örnekleme (SMOTE) yöntemleri kullanılmıştır. Sınıflandırma performansı üzerindeki etkiler, torbalama (bagging) ve arttırma (boosting) temelli kolektif öğrenme yöntem-lerine entegre edilmesiyle karşılaştırılmıştır. Algoritmaların doğru sınıflandırma performansları-nın karşılaştırılmasında doğruluk, Kappa istatistiği, duyarlılık ve seçicilik ölçütleri kullanılmıştır. Tüm istatistiksel analizler, açık kaynak kodlu bir yazılım olan R programlama dilinde yapılmıştır. Bulgular: Dengesiz veri setinde ham veri ile yapılan diyabet tanısı sınıflandırma başarısı oldukça düşüktür. Aşırı örnekleme yöntemi ile yapılan sınıflandırmaların, orijinal dengesiz veri seti, alt örnekleme ve sentetik azınlık aşırı örnekleme yöntemi ile yapılan sınıflandırmalardan çok daha başarılı tahmin gücüne sahip olduğu tespit edilmiştir. Sonuç: Sınıf dengesizliği varlığında veri setlerini yeniden örnekleme yöntemlerine tabi tutarak veriyi dengeledikten sonra sınıflandırma algoritmalarının kullanılması önerilmektedir. Anahtar Kelimeler: Kolektif öğrenme; sınıflandırma; dengesiz veri; hastalık tanısı; diyabet ABSTRACT Objective: Recently, machine learning methods have been widely used in disease diognosis. However, due to the large volume, multidimensional and complexity of the information , an unbalanced data problem arises. In this study, it is aimed to eliminate problem of imbalance by using re-sampling methods in an unbalanced data set related to diabetes patients, to classify diagnosis of diabetes with ensemble learning algorithms and to compare correct classification performances of algorithms. Materials and Methods: The data were collected from 185 patients older than 18 years of age who were admitted to Izmir Bozkaya Training and Research Hospital, Endocrinology and Metabolism Diseases outpatient clinic between June and September 2013. Under-sampling, over-sampling and synthetic minority over-sampling methods were used to eliminate unbalanced class problem for diagnosis of diabetes. The effects on classification performance were compared by integrating bagging and boosting methods into ensemble learning methods. Accuracy, Kappa statistics, sensitivity and specificity were used to compare correct classification performance of algorithms. All statistical analyzes were made in the R programming language, an open source software. Results: The success rate of diabetes diagnosis with raw data is very low in the unbalanced data set. It is determined that classifications made with over-sampling method have much more successful estimation power than classifications made with original unbalanced data set, under-sampling and synthetic minority over-sampling method. Conclusion: It is recommended to use classification algorithms after balancing the data by subjecting the data sets to resampling methods in the presence of class imbalance.

Class prediction models have been shown to have varying performances in clinical gene expression datasets. The accuracy of class prediction models differs from dataset to dataset and depends on the type of classification models. While a substantial amount of information is known about the characteristics of classification functions, little has been done to determine which characteristics of gene expression data have impact on the performance of a classifier. Gene expression is an important aspect of many genetic diseases in the context of genetic disorders as the disorder affects only few gene expressions. Therefore, gene-expression is important in identifying disease-gene associations. Hence this paper focuses on building a classification models that accurately predicts the Alzheimer's disease (AD) using machine learning technique called ensemble learning and feature selection algorithms to find out those few disease-gene-expressions.

We apply machine-learning techniques to construct linear, non-linear and ensemble machine learning algorithms. The linear comprises of logistic regression(LR), linear discriminant analysis(LDA) and least absolute shrinkage and selection operator (LASSO) while non-linear comprises of support vector machine(SVM), neural network(NN) and decision tree while ensemble learning are boosting, bagging and weighted average. We compared linear, non-linear and ensemble learning using the entire feature in consumer credit data sets. In the first experiment, the study revealed that ensemble learning performs significantly well follow by non-linear and linear machine learning. In the second experiment, we use the selected features from genetic algorithms. It was observed that there was a slight improvement in the result obtained from some of the linear and non-linear machine learning but slight decrease in the result obtained from ensemble learning. The grading scale was used on the overall accuracy of area under the curve(AUROC) in which bagging and weighted average score between 90-100 grade and performs excellently well, boosting obtained 80-90 grade, support vector machine obtained 70-80 grade, logistic regression, linear discriminant analysis, neural network and decision tree obtained 60-70 grade and lasso performs worse with 50-60 grade.

The schema matching problem consists of finding semantic correspondences between elements (e.g., attributes) of two database schemas. Typically, methods to solve this problem first use pair-wise functions called matchers to generate similarity scores values between pairs of elements from the two schemas. These scores are used as estimations of the correspondence between elements. Next, matchers are combined to establish which pairs of elements must be mapped when integrating the two schemas. For this, the best-known schema matching methods rely on fixed heuristics. We consider that using fixed heuristics is not always helpful to cope with a variety of database and element mismatch cases and argue that machine learning (ML) techniques comprise suitable alternatives to properly combine matchers. In this paper we propose ALMa (Active Learning Matching), a novel method for combining matchers based on active learning, which is an interesting and effective machine learning technique that efficiently exploits the users' expertise on the matching task. We report the results of experiments we carried out comparing ALMa with COMA, a well-known schema matching method in the literature based on fixed heuristics, and with YAM, a recent schema matching method based on supervised learning. In the experiments, ALMa achieved results better or at least similar to the baselines, while demanded less user effort, confirming the suitability of using active learning for combining matchers.

In this paper, we describe an approach to developing an ecologically valid framework for performing automated cognitive assessment. To automate assessment, we use a machine learning approach that builds a model of cognitive health based on observations of activity performance and uses lab-based assessment to provide ground truth for training and testing the learning algorithm. To evaluate our approach, we recruited older adults to perform a set of activities in our smart home test-bed. While participants perform activities, sensors placed in the smart home unobtrusively capture the progress of the activity. During analysis, we extract features that indicate how well participants perform the activities. Our machinelearning algorithm accepts these features as input and outputs the cognitive status of the participants as belonging to one of two groups: Cognitively healthy or Dementia. We conclude that machine-learning algorithms can distinguish between cognitively healthy older adults and older adults with dementia given adequate features that represent how well they have performed the activity.