Henry Joutsijoki

Kernel selection in multi-class support vector machines and its consequence to the number of ties in majority voting method

Support vector machines are a relatively new classification method which has nowadays established a firm foothold in the area of machine learning. It has been applied to numerous targets of applications. Automated taxa identification of benthic macroinvertebrates has got generally very little attention and especially using a support vector machine in it. In this paper we investigate how the changing of a kernel function in an SVM classifier effects classification results. A novel question is how the changing of a kernel function effects the number of ties in a majority voting method when we are dealing with a multi-class case. We repeated the classification tests with two different feature sets. Using SVM, we present accurate classification results proposing that SVM suits well to the automated taxa identification of benthic macroinvertebrates. We also present that the selection of a kernel has a great effect on the number of ties.

Evaluating the performance of artificial neural networks for the classification of freshwater benthic macroinvertebrates

Abstract Macroinvertebrates form an important functional component of aquatic ecosystems. Their ability to indicate various types of anthropogenic stressors is widely recognized which has made them an integral component of freshwater biomonitoring. The use of macroinvertebrates in biomonitoring is dependent on manual taxa identification which is currently a time-consuming and cost-intensive process conducted by highly trained taxonomical experts. Automated taxa identification of macroinvertebrates is a relatively recent research development. Previous studies have displayed great potential for solutions to this demanding data mining application. In this research we have a collection of 1350 images from eight different macroinvertebrate taxa and the aim is to examine the suitability of artificial neural networks (ANNs) for automated taxa identification of macroinvertebrates. More specifically, the focus is drawn on different training algorithms of Multi-Layer Perceptron (MLP), probabilistic neural network (PNN) and Radial Basis Function network (RBFN). We performed thorough experimental tests and we tested altogether 13 training algorithms for MLPs. The best classification accuracy of MLPs, 95.3%, was obtained by two conjugate gradient backpropagation variations and scaled conjugate gradient backpropagation. For PNN 92.8% and for RBFN 95.7% accuracies were achieved. The results show how important a proper choice of ANN is in order to obtain high accuracy in the automated taxa identification of macroinvertebrates and the obtained model can outperform the level of identification which is made by a taxonomist.

Comparing the one-vs-one and one-vs-all methods in benthic macroinvertebrate image classification

This paper investigates automated benthic macroinvertebrate identification and classification with multi-class support vector machines. Moreover, we examine, how the feature selection effects results, when one-vs-one and one-vsall methods are used. Lastly, we explore what happens for the number of tie situations with different kernel function selections. Our wide experimental tests with three feature sets and seven kernel functions indicated that one-vs-one method suits best for the automated benthic macroinvertebrate identification. In addition, we obtained clear differences to the number of tie situations with different kernel funtions. Furthermore, the feature selection had a clear influence on the results.

DAGSVM vs. DAGKNN: an experimental case study with benthic macroinvertebrate dataset

In this paper we examined the suitability of the Directed Acyclic Graph Support Vector Machine (DAGSVM) and Directed Acyclic Graph k-Nearest Neighbour (DAGKNN) method in classification of the benthic macroinvertebrate samples. We divided our 50 species dataset into five ten species groups according to their group sizes. We performed extensive experimental tests with every group, where DAGSVM was tested with seven kernel functions and DAGKNN with four measures. Feature selection was made by the scatter method [8]. Results showed that the quadratic and RBF kernel functions were the best ones and in the case of DAGKNN all measures produced quite similar results. Generally, the DAGSVM gained higher accuracies than DAGKNN, but still DAGKNN is a respectable option in benthic macroinvertebrate classification.

Machine Learning Approach to Automated Quality Identification of Human Induced Pluripotent Stem Cell Colony Images

The focus of this research is on automated identification of the quality of human induced pluripotent stem cell (iPSC) colony images. iPS cell technology is a contemporary method by which the patients cells are reprogrammed back to stem cells and are differentiated to any cell type wanted. iPS cell technology will be used in future to patient specific drug screening, disease modeling, and tissue repairing, for instance. However, there are technical challenges before iPS cell technology can be used in practice and one of them is quality control of growing iPSC colonies which is currently done manually but is unfeasible solution in large-scale cultures. The monitoring problem returns to image analysis and classification problem. In this paper, we tackle this problem using machine learning methods such as multiclass Support Vector Machines and several baseline methods together with Scaled Invariant Feature Transformation based features. We perform over 80 test arrangements and do a thorough parameter value search. The best accuracy (62.4%) for classification was obtained by using a k-NN classifier showing improved accuracy compared to earlier studies.

Prediction of a state of a subject on the basis of a stabilogram signal and video oculography test

Postural stability decreases with ageing and may lead to accidental falls, isolation and a reduction in the quality of life. The age at the onset of postural derangement, its extent and the reason for deterioration are poorly known within an individual, but in general it becomes more severe with age. In order to prevent falls and avoid severe injuries the postural derangement has to be noticed by the person and the possible nursing personnel. In this work we propose such numerical features, which can discriminate the persons having good or poor postural stability. These features can also be utilized to measure the outcome and progression of balance training. With these postural stability algorithms providing stability features for a subject we managed to classify correctly the type of stance on the force platform in more than 80% of sixty subjects. We used k-nearest neighbor algorithm as an intuitive baseline method and compared its results with those of support vector machines and hidden Markov models.

Error-Correcting Output Codes in Classification of Human Induced Pluripotent Stem Cell Colony Images

The purpose of this paper is to examine how well the human induced pluripotent stem cell (hiPSC) colony images can be classified using error-correcting output codes (ECOC). Our image dataset includes hiPSC colony images from three classes (bad, semigood, and good) which makes our classification task a multiclass problem. ECOC is a general framework to model multiclass classification problems. We focus on four different coding designs of ECOC and apply to each one of them k-Nearest Neighbor (k-NN) searching, naïve Bayes, classification tree, and discriminant analysis variants classifiers. We use Scaled Invariant Feature Transformation (SIFT) based features in classification. The best accuracy (62.4%) is obtained with ternary complete ECOC coding design and k-NN classifier (standardized Euclidean distance measure and inverse weighting). The best result is comparable with our earlier research. The quality identification of hiPSC colony images is an essential problem to be solved before hiPSCs can be used in practice in large-scale. ECOC methods examined are promising techniques for solving this challenging problem.

Homicide and Its Social Context: Analysis Using the Self-Organizing Map

Homicide is one of the most serious kinds of offenses. Research on causes of homicide has never reached a definite conclusion. The purpose of this article is to put homicide in its broad range of social context to seek correlation between this offense and other macroscopic socioeconomic factors. This international-level comparative study used a dataset covering 181 countries and 69 attributes. The data were processed by the Self-Organizing Map (SOM) assisted by other clustering methods, including ScatterCounter for attribute selection, and several statistical methods for obtaining comparable results. The SOM is found to be a useful tool for mapping criminal phenomena through processing of multivariate data, and correlation can be identified between homicide and socioeconomic factors.

Histogram-based classification of iPSC colony images using machine learning methods

This paper focuses on induced pluripotent stem cell (iPSC) colony image classification using machine learning methods and different feature sets obtained from the intensity histograms. Intensity histograms are obtained from the whole iPSC colony images and as a baseline for it they are determined only from the iPSC colony area of images. Furthermore, we apply to both of the datasets two simple feature selection methods having altogether four datasets. Altogether, 30 different classification methods are tested and we perform thorough experimental tests. The best accuracy (55%) is obtained for the feature set evaluated from the whole image using Directed Acyclic Graph Support Vector Machines (DAGSVM). DAGSVM is also the best choice when intensity histograms are evaluated only from the iPSC colony area. By this means accuracy of 54% is achieved. The obtained results are promising for further research where, for instance, more sophisticated feature selection and extraction methods and other multi-class extensions of SVM will be examined. However, intensity histograms are not alone adequate for iPSC colony image classification.

Investigating local spatially-enhanced structural and textural descriptors for classification of iPSC colony images

Induced pluripotent stem cells (iPSC) can be derived from fully differentiated cells of adult individuals and used to obtain any other cell type of the human body. This implies numerous prospective applications of iPSCs in regenerative medicine and drug development. In order to obtain valid cell culture, a quality control process must be applied to identify and discard abnormal iPSC colonies. Computer vision systems that analyze visual characteristics of iPSC colony health can be especially useful in automating and improving the quality control process. In this paper, we present an ongoing research that aims at the development of local spatially-enhanced descriptors for classification of iPSC colony images. For this, local oriented edges and local binary patterns are extracted from the detected colony regions and used to represent structural and textural properties of the colonies, respectively. We preliminary tested the proposed descriptors in classifying iPSCs colonies according to the degree of colony abnormality. The tests showed promising results for both, detection of iPSC colony borders and colony classification.