Data of different levels of complexity and of ever growing diversity of characteristics are the raw materials that machine learning practitioners try to model using their wide palette of methods and tools. The obtained models are meant to be a synthetic representation of the available, observed data that captures some of their intrinsic regularities or patterns. Therefore, the use of machine learning techniques for data analysis can be understood as a problem of pattern recognition or, more informally, of knowledge discovery and data mining. There exists a gap, though, between data modeling and knowledge extraction. Models, de- pending on the machine learning techniques employed, can be described in diverse ways but, in order to consider that some knowledge has been achieved from their description, we must take into account the human cog- nitive factor that any knowledge extraction process entails. These models as such can be rendered powerless unless they can be interpreted ,a nd the process of human interpretation follows rules that go well beyond techni- cal prowess. For this reason, interpretability is a paramount quality that machine learning methods should aim to achieve if they are to be applied in practice. This paper is a brief introduction to the special session on interpretable models in machine learning, organized as part of the 20 th European Symposium on Artificial Neural Networks, Computational In- telligence and Machine Learning. It includes a discussion on the several works accepted for the session, with an overview of the context of wider research on interpretability of machine learning models.

Making machine learning models interpretable

In real world data are often plagued by missing values which adversely affects the final outcome of the analysis based on such data. The missing values can be handled using various techniques like deletion or imputation. Of late, R has become one of the most preferred platform for carrying out data analysis, and its popularity is growing further. R provides various packages for handling missing values through imputation. The presence of multiple packages however, calls for an analysis of their comparative performance and examine their suitability for handling a given set of data. The performance of different R packages may differ for different datasets and may depend on the size of the dataset and richness of the missing values in the datasets. In this paper, the authors perform comparative study of the performance of the common R packages, namely VIM, MICE, MissForest, and HMISC, used for missing value imputation. The authors measured the performances of the said packages in terms of their imputation time, imputation efficiency and the effect on the variance. The imputation efficiency was measured in terms of the difference in predictive performance of a model built using original dataset vis-a-vis a dataset with imputed values. Similarly, the variance of the variables in the original dataset was compared that of corresponding variables in the imputed dataset. A missing value imputation package can be considered to be better if it consumes less imputation time and provides high imputation accuracy. Also in terms of variance, one would like to have the imputation package maintain the original variance of the variables. On analysing the four imputation packages on two datasets over three predictive algorithms–Logistic Regression, Support Vector Machines, and Artificial Neural Networks–it was observed that the performances varies depending on the size of the dataset, and the missing values present in them. The study highlights that certain missing value package used in conjunction with a given predictive algorithm provides better performance, which is again a function of the dataset characteristics.

Handling missing values: A study of popular imputation packages in R

Computational intelligence techniques for medical diagnosis and prognosis: Problems and current developments

Advances in medical technology have allowed for customized prognosis, diagnosis, and treatment regimens that utilize multiple heterogeneous data sources. Multiple kernel learning (MKL) is well suited for the integration of multiple high throughput data sources. MKL remains to be under-utilized by genomic researchers partly due to the lack of unified guidelines for its use, and benchmark genomic datasets. We provide three implementations of MKL in R. These methods are applied to simulated data to illustrate that MKL can select appropriate models. We also apply MKL to combine clinical information with miRNA gene expression data of ovarian cancer study into a single analysis. Lastly, we show that MKL can identify gene sets that are known to play a role in the prognostic prediction of 15 cancer types using gene expression data from The Cancer Genome Atlas, as well as, identify new gene sets for the future research. Multiple kernel learning coupled with modern optimization techniques provides a promising learning tool for building predictive models based on multi-source genomic data. MKL also provides an automated scheme for kernel prioritization and parameter tuning. The methods used in the paper are implemented as an R package called RMKL package, which is freely available for download through CRAN at https://CRAN.R-project.org/package=RMKL
 .

/pdf/multiple-kernel-learning-for-genomic-data-mining-and-2j6709bfwp.pdf

Multiple-kernel learning for genomic data mining and prediction

Multiple kernel learning (MKL) is a parametric kernel learning approach which allows the combination of multiple kernels for a given learning task. Studies reported in the literature have demonstrated the potentiality of MKL algorithms to address a wide range of machine learning tasks and especially biomedical applications. The aim of this paper is to present a review of MKL algorithms in order classification, feature selection and feature fusion problems to be addressed. Through the review the following issues are presented: a) the key properties of the MKL algorithms, b) how the MKL algorithms address issues regarding the nature of the datasets (missing data, multi classes, categorical features etc.), and c) the selection of kernels.

Multiple Kernel Learning Algorithms and Their Use in Biomedical Informatics

We discuss several approaches that make possible for kernel methods to deal with missing values. The first two are extended kernels able to handle missing values without data preprocessing methods. Another two methods are derived from a sophisticated multiple imputation technique involving logistic regression as local model learner. The performance of these approaches is compared using a binary data set that arises typically in microbiology (the microbial source tracking problem). Our results show that the kernel extensions demonstrate competitive performance in comparison with multiple imputation in terms of predictive accuracy. However, these results are achieved with a simpler and deterministic methodology and entail a much lower computational effort.

Handling missing values in kernel methods with application to microbiology data

A two-layer neural network is developed in which the neu- ron model computes a user-defined similarity function between inputs and weights. The neuron model is formed by the composition of an adapted lo- gistic function with the mean of the partial input-weight similarities. The model is capable of dealing directly with variables of potentially different nature (continuous, ordinal, categorical); there is also provision for missing values. The network is trained using a fast two-stage procedure and in- volves the setting of only one parameter. In our experiments, the network achieves slightly superior performance on a set of challenging problems with respect to both RBF nets and RBF-kernel SVMs.

/pdf/similarity-networks-for-heterogeneous-data-29hpwmf5ry.pdf

Similarity networks for heterogeneous data

In kernel-based machines, the integration of several kernels to build more flexible learning methods is a promising avenue for research. In particular, in Multiple Kernel Learning a compound kernel is build by learning a kernel that is the weighted mean of several sources. We show in this paper that the only feasible average for kernel learning is precisely the arithmetic average. We also show that three familiar means (the geometric, inverse root mean square and harmonic means) for positive
real values actually generate valid kernels.

https://www.elen.ucl.ac.be/Proceedings/esann/esannpdf/es2012-60.pdf

Averaging of kernel functions

Non-negative Matrix Factorisation (NMF) has become a standard method for source identification when data, sources and mixing coefficients are constrained to be positive-valued. The method has recently been extended to allow for negative-valued data and sources in the form of Semi-and Convex-NMF. In this paper, we re-elaborate Semi-NMF within a full Bayesian framework. This provides solid foundations for parameter estimation and, importantly, a principled method to address the problem of choosing the most adequate number of sources to describe the observed data. The proposed Bayesian Semi-NMF is preliminarily evaluated here in a real neuro-oncology problem.

https://www.elen.ucl.ac.be/Proceedings/esann/esannpdf/es2016-62.pdf

Bayesian semi non-negative matrix factorisation

We present a novel method that aims at providing a more stable selection of feature subsets when variations in the training process occur. This is accomplished by using an instance-weighting process -assigning different importances to instances as a preprocessing step to a feature weighting method that is independent of the learner, and then making good use of both sets of computed weigths in a standard Nearest-Neighbours classifier.
We report extensive experimentation in well-known benchmarking datasets as well as some challenging microarray
gene expression problems. Our results show increases in stability for most subset sizes and most problems, without
compromising prediction accuracy.

Luis Antonio Belanche Muñoz

Papers

Handling missing values in kernel methods with application to microbiology data

Similarity networks for heterogeneous data

Averaging of kernel functions

Bayesian semi non-negative matrix factorisation

Instance and feature weighted k-nearest-neighbors algorithm