Search

In the present state of technology, rotating machine health monitoring is becoming economically mandatory. There is presently a clear trend to apply deep learning techniques to this area. This work analyzes some of the main supervised and unsupervised neural network current architectures and, as a consequence, proposes a novel denoising convolutional autoencoder model configuration applied to machine health monitoring. The configuration aims the main advantages of different conventional architectures for performance optimization over normalized time series signals. Experiments carried out based on vibration data extracted from bearings helped to validate the developed model and to determine the structure that offer the most satisfactory results. A comparison is made in terms of accuracy and robustness to noise between the proposed method and some related methods, including other published models. The objective is to assess the effectiveness of the proposed method through using two public experimental datasets produced under different machine operating conditions. The experimental results show that the proposed architecture provides greater classification accuracy and robustness in both case studies considering other published models.

Precision medicine in oncology aims at obtaining data from heterogeneous sources to have a precise estimation of a given patient's state and prognosis. With the purpose of advancing to personalized medicine framework, accurate diagnoses allow prescription of more effective treatments adapted to the specificities of each individual case. In the last years, next-generation sequencing has impelled cancer research by providing physicians with an overwhelming amount of gene-expression data from RNA-seq high-throughput platforms. In this scenario, data mining and machine learning techniques have widely contribute to gene-expression data analysis by supplying computational models to supporting decision-making on real-world data. Nevertheless, existing public gene-expression databases are characterized by the unfavorable imbalance between the huge number of genes (in the order of tenths of thousands) and the small number of samples (in the order of a few hundreds) available. Despite diverse feature selection and extraction strategies have been traditionally applied to surpass derived over-fitting issues, the efficacy of standard machine learning pipelines is far from being satisfactory for the prediction of relevant clinical outcomes like follow-up end-points or patient's survival. Using the public Pan-Cancer dataset, in this study we pre-train convolutional neural network architectures for survival prediction on a subset composed of thousands of gene-expression samples from thirty-one tumor types. The resulting architectures are subsequently fine-tuned to predict lung cancer progression-free interval. The application of convolutional networks to gene-expression data has many limitations, derived from the unstructured nature of these data. In this work we propose a methodology to rearrange RNA-seq data by transforming RNA-seq samples into gene-expression images, from which convolutional networks can extract high-level features. As an additional objective, we investigate whether leveraging the information extracted from other tumor-type samples contributes to the extraction of high-level features that improve lung cancer progression prediction, compared to other machine learning approaches.

This paper proposes an ensemble of forecasting methods based on neural networks/recurrent neural networks (E-LSTM). The aim of the algorithm is to help organizing the planting cycle using short-term rainfall forecasts when the data are taken from a single observation point. The computational models are carried out for univariate rainfall time series by of multi-step prediction horizons in combination of nonlinear autoregressive models (NAR) modified by several approaches such energy associated to series, subsampling methods and their combinations, which are heuristically modified by Bayesian inference and statistical roughness in the learning process. The study analyses and compares the relative advantages and limitations of each algorithm against the aforementioned to forecast rainfall from 1 to 6 months ahead. Simulation results illustrate the effectiveness of the E-LSTM approach through different series classified by their statistical roughness in both, the learning process and the validation test using the SMAPE and RMSE metrics. Comparisons also are made by adding fractional Gaussian noise to highlight the performance and constraints of the ensemble approach.

This article proposes that the combination of smoothing approach considering the entropic information provided by Renyi's method, has an acceptable performance in term of forecasting errors. The methodology of the proposed scheme is examined through benchmark chaotic time series, such as Mackey Glass, Lorenz, Henon maps, the Lynx and rainfall from Santa Francisca - Cordoba, with addition of white noise by using neural networks-based energy associated (EAS) predictor filter modified by Renyi's entropy of the series. When the time series is short or long, the underlying dynamical system is nonlinear and temporal dependencies span long time intervals, in which this are also called long memory process. In such cases, the inherent nonlinearity of neural networks models and a higher robustness to noise seem to partially explain their better prediction performance when entropic information is extracted from the series. Then, to demonstrate that permutation entropy is computationally efficient, robust to outliers, and effective to measure complexity of time series, computational results are evaluated against several non-linear ANN predictors to show the predictability of noisy rainfall and chaotic time series reported in the literature.

TNM staging system remains the main prognostic classification in many adult cancers in order to predict treatment outcomes.Machine learning algorithms are performed to find different patient groups in the treatment of breast cancer disease.The new approaches found novel group risks that improve prediction results of breast cancer relapse. Discretization of continuous variables is a common practice in medical research to identify risk patient groups. This work compares the performance of gold-standard categorization procedures (TNM+A protocol) with that of three supervised discretization methods from Machine Learning (CAIM, ChiM and DTree) in the stratification of patients with breast cancer. The performance for the discretization algorithms was evaluated based on the results obtained after applying standard survival analysis procedures such as Kaplan-Meier curves, Cox regression and predictive modelling. The results show that the application of alternative discretization algorithms could lead the clinicians to get valuable information for the diagnosis and outcome of the disease. Patient data were collected from the Medical Oncology Service of the Hospital Clinico Universitario (Malaga, Spain) considering a follow up period from 1982 to 2008.

The diagnosis and prognosis of cancer are among the more critical challenges that modern medicine confronts. In this sense, personalized medicine aims to use data from heterogeneous sources to estimate the evolution of the disease for each specific patient in order to fit the more appropriate treatments. In recent years, DNA sequencing data have boosted cancer prediction and treatment by supplying genetic information that has been used to design genetic signatures or biomarkers that led to a better classification of the different subtypes of cancer as well as to a better estimation of the evolution of the disease and the response to diverse treatments. Several machine learning models have been proposed in the literature for cancer prediction. However, the efficacy of these models can be seriously affected by the existing imbalance between the high dimensionality of the gene expression feature sets and the number of samples available, what is known as the curse of dimensionality. Although linear predictive models could give worse performance rates when compared to more sophisticated non-linear models, they have the main advantage of being interpretable. However, the use of domain-specific information has been proved useful to boost the performance of multivariate linear predictors in high dimensional settings. In this work, we design a set of linear predictive models that incorporate domain-specific information from genetic pathways for effective feature selection. By combining these linear model with other classical machine learning models, we get state-of-art performance rates in the prediction of vital status on a public cancer dataset.

The diagnosis and prognosis of cancer are among the more challenging tasks that oncology medicine deals with. With the main aim of fitting the more appropriate treatments, current personalized medicine focuses on using data from heterogeneous sources to estimate the evolution of a given disease for the particular case of a certain patient. In recent years, next-generation sequencing data have boosted cancer prediction by supplying gene-expression information that has allowed diverse machine learning algorithms to supply valuable solutions to the problem of cancer subtype classification, which has surely contributed to better estimation of patient’s response to diverse treatments. However, the efficacy of these models is seriously affected by the existing imbalance between the high dimensionality of the gene expression feature sets and the number of samples available for a particular cancer type. To counteract what is known as the curse of dimensionality, feature selection and extraction methods have been traditionally applied to reduce the number of input variables present in gene expression datasets. Although these techniques work by scaling down the input feature space, the prediction performance of traditional machine learning pipelines using these feature reduction strategies remains moderate. In this work, we propose the use of the Pan-Cancer dataset to pre-train deep autoencoder architectures on a subset composed of thousands of gene expression samples of very diverse tumor types. The resulting architectures are subsequently fine-tuned on a collection of specific breast cancer samples. This transfer-learning approach aims at combining supervised and unsupervised deep learning models with traditional machine learning classification algorithms to tackle the problem of breast tumor intrinsic-subtype classification. Our main goal is to investigate whether leveraging the information extracted from a large collection of gene expression data of diverse tumor types contributes to the extraction of useful latent features that ease solving a complex prediction task on a specific neoplasia.

Data augmentation has been proven particularly effective for image classification tasks where a significant boost of prediction accuracy can be obtained when the technique is combined with the use of Deep Learning architectures. Unfortunately, for non-image data the situation is quite different and the positive effect of augmenting the training set size is much smaller. In this work, we propose a method that creates new samples by adjusting the level of noise for individual input variables previously ranked by their relevance level. Results from several tests are analyzed using nine benchmark data sets when the augmented and original data are used for supervised training on Deep Learning architectures.

We propose a new ensemble forecasting method, Bayesian enhanced ensemble approach (BEEA), to model neural networks for time series forecasting. Motivated by looking to get better prediction algorithms, our proposed algorithm can effectively integrate the data from different methods modified by Bayesian learning. A dataset of univariate time series (seasonal and non-seasonal) is used to forecast short-long prediction horizons, mainly 3 and 18 out-of-sample. Comparative simulations of our method with ten existing linear and nonlinear forecasting approaches, namely, Energy associated to series (EAS), Energy associated to series modify by Renyis’s entropy (EASmod), Bayesian approach (BA), Bayesian enhanced (BEA), Bayesian enhanced modified by Renyi’s entropy (BEMA), Bayesian enhanced modified by relative entropy (BEAmod.) is presented. Relative advantages and limitations of the ensemble approach in contrast with some reported in the literature illustrates the effectiveness of the Bayesian enhanced ensemble approach (BEEA) through different forecasting horizons in both, the learning process and the validation test using the MASE, SMAPE and RMSE forecast error metrics to highlight the performance and limitations of the BEEA approach.

Clinicians are interested in the estimation of robust and relevant genetic signatures from gene sequencing data. Many machine learning approaches have been proposed trying to address well-known issues of this complex task (feature or gene selection, classification or model selection, and prediction assessment). Addressing this problem often requires a deep knowledge of these methods and some of them demand high computational resources that may not be affordable. In this paper, an exhaustive study that includes different types of feature selection methods and classifiers is presented, providing clinicians an useful insight of the most suitable methods for this purpose. Predictions assessment is performed using a bootstrap crossvalidation strategy as an honest validation scheme. The results of this study for six benchmark datasets show that filter or embedded methods are preferred, in general, to wrapper methods according to their better statistical significant results, in terms of accuracy, and lower demand for computational resources. Universidad de Málaga. Campus de Excelencia Internacional Andalucía Tech.

Biomarkers discovery in RNA-Seq gene expression data aims to identify a subset of genes with high prediction capabilities in complex human traits. Ideally, robustness of a genetic signature is an a priori desired property difficult to obtain in practice due to the nature of the data. In this paper we propose a simple way of including biological knowledge into an \(l_1\)-regularization model. In particular, the paper describes two different approaches (Gene-specific, Gene-disease) of how to extract biological information from an online resource (PubTator) based on a gene citations scheme, and it shows how to use it to estimate a predictive model. The results of the analysis carried out in a public RNA-Seq dataset of breast invasive cancer (BRCA) shows that our proposal obtains some improvement in the predictive performance and a substantial improvement in terms of biomarkers stability (genetic signatures 400% times more robust in the worst case scenario).

Urda, D., Franco, L. and Jerez, J.M. (2017). Classification of high dimensional data using LASSO ensembles. Proceedings IEEE SSCI'17, Symposium Series on Computational Intelligence, Honolulu, Hawaii, U.S.A. (2017). ISBN: 978-1-5386-2726-6 The estimation of multivariable predictors with good performance in high dimensional settings is a crucial task in biomedical contexts. Usually, solutions based on the application of a single machine learning model are provided while the use of ensemble methods is often overlooked within this area despite the well-known benefits that these methods provide in terms of predictive performance. In this paper, four ensemble approaches are described using LASSO base learners to predict the vital status of a patient from RNA-Seq gene expression data. The results of the analysis carried out in a public breast invasive cancer (BRCA) dataset shows that the ensemble approaches outperform statistically significant the standard LASSO model considered as baseline case. We also perform an analysis of the computational costs involved for each of the approaches, providing different usage recommendations according to the available computational power. Universidad de Málaga. Campus de Excelencia Internacional Andalucía Tech

This paper propose that the combination of smoothing approach taking into account the entropic information provided by Renyi' method, has an acceptable performance in term of forecasting errors. The methodology of the proposed scheme is examined through benchmark chaotic time series, such as Mackay Glass, Lorenz, Henon maps, the Lynx and rainfall from Santa Francisca series, with addition of white noise by using neural networks-based energy associated (EAS) predictor filter modified by Renyi entropy of the series. In particular, when the time series is short or long, the underlying dynamical system is nonlinear and temporal dependencies span long time intervals, in which this are also called long memory process. In such cases, the inherent nonlinearity of neural networks models and a higher robustness to noise seem to partially explain their better prediction performance when entropic information is extracted from the series. Then, to demonstrate that permutation entropy is computationally efficient, robust to outliers, and effective to measure complexity of time series, computational results are evaluated against several non-linear ANN predictors proposed before to show the predictability of noisy rainfall and chaotic time series reported in the literature.

One of the prevailing applications of machine learning is the use of predictive modelling in clinical survival analysis. In this work, we present our view of the current situation of computer tools for survival analysis, stressing the need of transferring the latest results in the field of machine learning to biomedical researchers. We propose a web based software for survival analysis called OSA (Online Survival Analysis), which has been developed as an open access and user friendly option to obtain discrete time, predictive survival models at individual level using machine learning techniques, and to perform standard survival analysis. OSA employs an Artificial Neural Network (ANN) based method to produce the predictive survival models. Additionally, the software can easily generate survival and hazard curves with multiple options to personalise the plots, obtain contingency tables from the uploaded data to perform different tests, and fit a Cox regression model from a number of predictor variables. In the Materials and Methods section, we depict the general architecture of the application and introduce the mathematical background of each of the implemented methods. The study concludes with examples of use showing the results obtained with public datasets., This work was supported by grants TIN2010-16556 from MICINN-SPAIN (Spanish Government) and P08-TIC-4026 (Andalusia Regional Government, Spain). All of them include FEDER funds (European Union). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Objectives: Missing data imputation is an important task in cases where it is crucial to use all available data and not discard records with missing values. This work evaluates the performance of several statistical and machine learning imputation methods that were used to predict recurrence in patients in an extensive real breast cancer data set. Materials and methods: Imputation methods based on statistical techniques, e.g., mean, hot-deck and multiple imputation, and machine learning techniques, e.g., multi-layer perceptron (MLP), self-organisation maps (SOM) and k-nearest neighbour (KNN), were applied to data collected through the ''El Alamo-I'' project, and the results were then compared to those obtained from the listwise deletion (LD) imputation method. The database includes demographic, therapeutic and recurrence-survival information from 3679 women with operable invasive breast cancer diagnosed in 32 different hospitals belonging to the Spanish Breast Cancer Research Group (GEICAM). The accuracies of predictions on early cancer relapse were measured using artificial neural networks (ANNs), in which different ANNs were estimated using the data sets with imputed missing values. Results: The imputation methods based on machine learning algorithms outperformed imputation statistical methods in the prediction of patient outcome. Friedman's test revealed a significant difference (p=0.0091) in the observed area under the ROC curve (AUC) values, and the pairwise comparison test showed that the AUCs for MLP, KNN and SOM were significantly higher (p=0.0053, p=0.0048 and p=0.0071, respectively) than the AUC from the LD-based prognosis model. Conclusion: The methods based on machine learning techniques were the most suited for the imputation of missing values and led to a significant enhancement of prognosis accuracy compared to imputation methods based on statistical procedures.

Background: Extracting relevant information from microarray data is a very complex task due to the characteristics of the data sets, as they comprise a large number of features while few samples are generally available. In this sense, feature selection is a very important aspect of the analysis helping in the tasks of identifying relevant genes and also for maximizing predictive information. Methods: Due to its simplicity and speed, Stepwise Forward Selection (SFS) is a widely used feature selection technique. In this work, we carry a comparative study of SFS and Genetic Algorithms (GA) as general frameworks for the analysis of microarray data with the aim of identifying group of genes with high predictive capability and biological relevance. Six standard and machine learning-based techniques (Linear Discriminant Analysis (LDA), Support Vector Machines (SVM), Naive Bayes (NB), C-MANTEC Constructive Neural Network, K-Nearest Neighbors (kNN) and Multilayer perceptron (MLP)) are used within both frameworks using six free-public datasets for the task of predicting cancer outcome. Results: Better cancer outcome prediction results were obtained using the GA framework noting that this approach, in comparison to the SFS one, leads to a larger selection set, uses a large number of comparison between genetic profiles and thus it is computationally more intensive. Also the GA framework permitted to obtain a set of genes that can be considered to be more biologically relevant. Regarding the different classifiers used standard feedforward neural networks (MLP), LDA and SVM lead to similar and best results, while C-MANTEC and k-NN followed closely but with a lower accuracy. Further, C-MANTEC, MLP and LDA permitted to obtain a more limited set of genes in comparison to SVM, NB and kNN, and in particular C-MANTEC resulted in the most robust classifier in terms of changes in the parameter settings. Conclusions: This study shows that if prediction accuracy is the objective, the GAbased approach lead to better results respect to the SFS approach, independently of the classifier used. Regarding classifiers, even if C-MANTEC did not achieve the best overall results, the performance was competitive with a very robust behaviour in terms of the parameters of the algorithm, and thus it can be considered as a candidate technique for future studies.

Display Omitted A novel two-stage procedure to find genetic profiles is presented.Biological information and an evolutionary strategy are combined.Biological information improves robustness and prognosis.Better accuracy results are obtained over the three datasets analyzed. Genetic algorithms are widely used in the estimation of expression profiles from microarrays data. However, these techniques are unable to produce stable and robust solutions suitable to use in clinical and biomedical studies. This paper presents a novel two-stage evolutionary strategy for gene feature selection combining the genetic algorithm with biological information extracted from the KEGG database. A comparative study is carried out over public data from three different types of cancer (leukemia, lung cancer and prostate cancer). Even though the analyses only use features having KEGG information, the results demonstrate that this two-stage evolutionary strategy increased the consistency, robustness and accuracy of a blind discrimination among relapsed and healthy individuals. Therefore, this approach could facilitate the definition of gene signatures for the clinical prognosis and diagnostic of cancer diseases in a near future. Additionally, it could also be used for biological knowledge discovery about the studied disease.

Data discretization is an important processing step for several computational methods that work only with binary input data. In this work a method for discretize continuous data based on the use of the Extreme Learning Machine neural network architecture is developed and tested. The new method does not use data labels for performing the discretization process and thus is suitable for supervised and supervised data and also, as it is based on the Extreme Learning Machine, is very fast even for large input data sets. The efficiency of the new method is analyzed on several benchmark functions, testing the classification accuracy obtained with raw and discretized data, and also in comparison to results from the application of a state-of-the-art supervised discretization algorithm. The results indicate the suitability of the developed approach.

Microarray data analysis is attracting increasing attention in computer science because of the many applications of machine learning methods in prediction problems. The process typically involves a feature selection step, important in order to increase the accuracy and speed of the classifiers. This work analyzes the characteristics of the features selected by two wrapper methods, the first one based on artificial neural networks (ANN) and the second in a novel constructive neural network (CNN) algorithm, to later propose a hybrid model that combines the advantages of wrapper and filter methods. The results obtained in terms of the computational costs involved and the prediction accuracy reached show the feasibility of the hybrid model proposed here and indicate an interesting research line for the near future.

This paper studies the extension of the Generalization Complexity (GC) measure to real valued input problems The GC measure, defined in Boolean space, was proposed as a simple tool to estimate the generalization ability that can be obtained when a data set is learnt by a neural network Using two different discretization methods, the real valued inputs are transformed into binary values, from which the generalization complexity can be straightforwardly computed The discretization transformation is carried out both through a very simple method based on equal width intervals (EW) and with a more sophisticated supervised method (the CAIM algorithm) that use much more information about the data A study of the relationship between data complexity and generalization ability obtained was done together with an analysis of the relationship between best neural architecture size and complexity.

This chapter presents and discusses several well-known constructive neural network algorithms suitable for constructing feedforward architectures aiming at classification tasks involving two classes. The algorithms are divided into two different groups: the ones directed by the minimization of classification errors and those based on a sequential model. In spite of the focus being on two-class classification algorithms, the chapter also briefly comments on the multiclass versions of several two-class algorithms, highlights some of the most popular constructive algorithms for regression problems and refers to several other alternative algorithms.

Constructive neural network algorithms suffer severely from overfitting noisy datasets as, in general, they learn the set of examples until zero error is achieved. We introduce in this work a method for detect and filter noisy examples using a recently proposed constructive neural network algorithm. The method works by exploiting the fact that noisy examples are harder to be learnt, needing a larger number of synaptic weight modifications than normal examples. Different tests are carried out, both with controlled experiments and real benchmark datasets, showing the effectiveness of the approach.

This paper presents an evolved rule-based tool for mammography interpretation denominated "COBRA: Catalonia online breastcancer risk assessor." COBRA is designed to aid radiologists in the interpretation of mammography to decide whether to perform a biopsy on a patient or not while providing a human-friendly explanation of the underlying reasoning. From a diagnostic point of view, the tool exhibits high performance measures (i.e., sensitivity, specificity, and positive predictive value). From an interpretability point of view, COBRA's behavior is explained by only 14 rules containing, in average, 2.73 conditions perrule.

Several approaches have been proposed for the analysis of DNA microarray datasets, focusing on the performance and robustness of the final feature subsets. The novelty of this paper arises in the use of q-values to pre-filter the features of a DNA microarray dataset identifying the most significant ones and including this information into a genetic algorithm for further feature selection. This method is applied to a lung cancer microarray dataset resulting in similar performance rates and greater robustness in terms of selected features (on average a 36.21% of robustness improvement) when compared to results of the standard algorithm.