Feature selection is an important component of many machine learning applications. Especially in many bioinformatics tasks, efficient and robust feature selection methods are desired to extract meaningful features and eliminate noisy ones. In this paper, we propose a new robust feature selection method with emphasizing joint l2,1-norm minimization on both loss function and regularization. The l2,1-norm based loss function is robust to outliers in data points and the l2,1-norm regularization selects features across all data points with joint sparsity. An efficient algorithm is introduced with proved convergence. Our regression based objective makes the feature selection process more efficient. Our method has been applied into both genomic and proteomic biomarkers discovery. Extensive empirical studies are performed on six data sets to demonstrate the performance of our feature selection method.

/pdf/efficient-and-robust-feature-selection-via-joint-l2-1-norms-kgrf4q2v0t.pdf

Efficient and Robust Feature Selection via Joint ℓ2,1-Norms Minimization

C4.5: Programs for Machine Learning (書評)

Knowledge Discovery in Databases (KDD) is an active and important research area with the promise for a high payoff in many business and scientific applications. One of the main tasks in KDD is classification. A particular efficient method for classification is decision tree induction. The selection of the attribute used at each node of the tree to split the data (split criterion) is crucial in order to correctly classify objects. Different split criteria were proposed in the literature (Information Gain, Gini Index, etc.). It is not obvious which of them will produce the best decision tree for a given data set. A large amount of empirical tests were conducted in order to answer this question. No conclusive results were found. In this paper we introduce a formal methodology, which allows us to compare multiple split criteria. This permits us to present fundamental insights into the decision process. Furthermore, we are able to present a formal description of how to select between split criteria for a given data set. As an illustration we apply the methodology to two widely used split criteria: Gini Index and Information Gain.

Theoretical Comparison between the Gini Index and Information Gain Criteria

http://csmres.co.uk/cs.public.upd/article-downloads/Machine-learning-approaches-in-drug-discovery-methods-and-applications.pdf

Machine-learning approaches in drug discovery: methods and applications.

With the technology development in cyber physical systems and big data, there are huge potential to apply them to achieve personalization and improve resource efficiency in Industry 4.0. As Industr...

Big data for cyber physical systems in industry 4.0: a survey

Decision tree induction has become one of the most popular methods for classification and prediction. The key step in the process of inferring decision trees is finding the right criteria for splitting the training set into smaller and smaller subsets so that, ideally, all elements of a subset finally belong to one class. These split criteria can be defined in different ways (e.g. minimizing impurity of a subset, or minimizing entropy in a subset), and therefore they emphasize different properties of the inferred tree, such as size or classification accuracy. In this paper we analyze if the split functions introduced in a statistical and machine learning context are also well suited for a KDD context. We selected two well known split functions, namely Gini Index (CART) and Information Gain (C4.5) and introduced our own family of split functions and tested them on 9,000 data sets of different sizes (from 200 to 20,000 tuples). The tests have shown that the two popular functions are very sensitive to the variation of the training set sizes and therefore the quality of the inferred trees is highly dependent on the training set size. At the same time however, we were able to show that the simplest members of the introduced family of split functions behave in a very predictable way and, furthermore, the created trees were superior to the trees inferred using the Gini Index or the Information Gain based on our evaluation criteria.

Selecting Optimal Split-Functions for Large Datasets

Assessment of the universality of the electrophysiological signal acquired from tomatoes and eggplants

Plant electrophysiology carries a strong potential for assessing the health of a plant. Current literature for the classification of plant electrophysiology generally comprises classical methods based on signal features that portray a simplification of the raw data and introduce a high computational cost. The Deep Learning (DL) techniques automatically learn the classification targets from the input data, overcoming the need for precalculated features. However, they are scarcely explored for identifying plant stress on electrophysiological recordings. This study applies DL techniques to the raw electrophysiological data from 16 tomato plants growing in typical production conditions to detect the presence of stress caused by a nitrogen deficiency. The proposed approach predicts the stressed state with an accuracy of around 88%, which could be increased to over 96% using a combination of the obtained prediction confidences. It outperforms the current state-of-the-art with over 8% higher accuracy and a potential for a direct application in production conditions. Moreover, the proposed approach demonstrates the ability to detect the presence of stress at its early stage. Overall, the presented findings suggest new means to automatize and improve agricultural practices with the aim of sustainability. 

/pdf/detecting-stress-caused-by-nitrogen-deficit-using-deep-2ghk0m80.pdf

Laura Elena Raileanu

Papers

Theoretical Comparison between the Gini Index and Information Gain Criteria

Selecting Optimal Split-Functions for Large Datasets

Assessment of the universality of the electrophysiological signal acquired from tomatoes and eggplants

Detecting stress caused by nitrogen deficit using deep learning techniques applied on plant electrophysiological data