Oded Maimon

Top-down induction of decision trees classifiers - a survey

Decision trees are considered to be one of the most popular approaches for representing classifiers. Researchers from various disciplines such as statistics, machine learning, pattern recognition, and data mining considered the issue of growing a decision tree from available data. This paper presents an updated survey of current methods for constructing decision tree classifiers in a top-down manner. The paper suggests a unified algorithmic framework for presenting these algorithms and describes the various splitting criteria and pruning methodologies.

Dynamic Scheduling and Routing for Flexible Manufacturing Systems that Have Unreliable Machines

This paper presents a method for real-time scheduling and routing of material in a Flexible Manufacturing System FMS. It extends the earlier scheduling work of Kimemia and Gershwin in which the FMS model includes machines that fail at random times and stay down for random lengths of time. The new element is the capability of different machines to perform some of the same operations. The times that different machines require to perform the same operation may differ. This paper includes a model, its analysis, a real-time algorithm, and examples.

The design process: properties, paradigms, and structure

In this paper, we examine the logic and methodology of engineering design from the perspective of the philosophy of science. The fundamental characteristics of design problems and design processes are discussed and analyzed. These characteristics establish the framework within which different design paradigms are examined. Following the discussions on descriptive properties of design, and the prescriptive role of design paradigms, we advocate the plausible hypothesis that there is a direct resemblance between the structure of design processes and the problem solving of scientific communities. The scientific community metaphor has been useful in guiding the development of general purpose highly effective design process meta-tools.

Information-theoretic algorithm for feature selection

Abstract Feature selection is used to improve the efficiency of learning algorithms by finding an optimal subset of features. However, most feature selection techniques can handle only certain types of data. Additional limitations of existing methods include intensive computational requirements and inability to identify redundant variables. In this paper, we present a novel, information-theoretic algorithm for feature selection, which finds an optimal set of attributes by removing both irrelevant and redundant features. The algorithm has a polynomial computational complexity and is applicable to datasets of a mixed nature. The method performance is evaluated on several benchmark datasets by using a standard classifier (C4.5).

Predictive Toxicology of Cobalt Nanoparticles and Ions: Comparative In Vitro Study of Different Cellular Models Using Methods of Knowledge Discovery from Data

The toxicological effects of cobalt nanoparticles (Co-NPs) aggregates were examined and compared with those of cobalt ions (Co-ions) using six different cell lines representing lung, liver, kidney, intestine, and the immune system. Dose-response curves were studied in the concentration range of 0.05-1.0 mM, employing 3-(4,5-dimethylthiazol-2-Yl)-2,5-diphenyltetrazolium bromide test, neutral red, and Alamar blue as end point assays following exposures for 48 and 72 h. Data analysis and predictive modeling of the obtained data sets were executed by employing a decision tree model (J48), where training and validation were carried out by an iterative process. It was established, as expected, that concentration is the highest rank parameter. This is because concentration parameter provides the highest information gain with respect to toxicity. The second-rank parameter emerged to be either the compound type (Co-ions or Co-NPs) or the cell model, depending on the concentration range. The third and the lowest rank in the model was exposure duration. The hierarchy of cell sensitivity toward cobalt ions was found to obey the following sequence of cell lines: A549 > MDCK > NCIH441 > Caco-2 > HepG2 > dendritic cells (DCs), with A549 being the most sensitive cell line and primary DCs were the least sensitive ones. However, a different hierarchy pattern emerged for Co-NPs: A549 = MDCK = NCIH441 = Caco-2 > DCs > HepG2. The overall findings are in line with the hypothesis that the toxic effects of aggregated cobalt NPs are mainly due to cobalt ion dissolution from the aggregated NPs.

Group set-up for printed circuit board assembly

The current practice in the assembly of electronic components on printed circuit boards (PCBs) is serial production, a process characterized by very long set-up times. However, with the advent of efficient on-line process information, new production control methods are now possible. This paper proposes a different production method, called the group set-up (GSU) method, which can significantly reduce set-up times. The traditional and the GSU production methods are compared, and it is shown that the GSU always performs better than the traditional method in terms of total production flow (throughput) and labour time. However, the traditional method performs better than the GSU in terms of work in process (WIP) inventory; and in some cases, in terms of makespan (lead time). A detailed analysis for a small number of PCBs is presented.

A compact and accurate model for classification

We describe and evaluate an information-theoretic algorithm for data-driven induction of classification models based on a minimal subset of available features. The relationship between input (predictive) features and the target (classification) attribute is modeled by a tree-like structure termed an information network (IN). Unlike other decision-tree models, the information network uses the same input attribute across the nodes of a given layer (level). The input attributes are selected incrementally by the algorithm to maximize a global decrease in the conditional entropy of the target attribute. We are using the prepruning approach: when no attribute causes a statistically significant decrease in the entropy, the network construction is stopped. The algorithm is shown empirically to produce much more compact models than other methods of decision-tree learning while preserving nearly the same level of classification accuracy.

Privacy-preserving data mining: A feature set partitioning approach

In privacy-preserving data mining (PPDM), a widely used method for achieving data mining goals while preserving privacy is based on k-anonymity. This method, which protects subject-specific sensitive data by anonymizing it before it is released for data mining, demands that every tuple in the released table should be indistinguishable from no fewer than k subjects. The most common approach for achieving compliance with k-anonymity is to replace certain values with less specific but semantically consistent values. In this paper we propose a different approach for achieving k-anonymity by partitioning the original dataset into several projections such that each one of them adheres to k-anonymity. Moreover, any attempt to rejoin the projections, results in a table that still complies with k-anonymity. A classifier is trained on each projection and subsequently, an unlabelled instance is classified by combining the classifications of all classifiers. Guided by classification accuracy and k-anonymity constraints, the proposed data mining privacy by decomposition (DMPD) algorithm uses a genetic algorithm to search for optimal feature set partitioning. Ten separate datasets were evaluated with DMPD in order to compare its classification performance with other k-anonymity-based methods. The results suggest that DMPD performs better than existing k-anonymity-based algorithms and there is no necessity for applying domain dependent knowledge. Using multiobjective optimization methods, we also examine the tradeoff between the two conflicting objectives in PPDM: privacy and predictive performance.

Soft Computing for Knowledge Discovery and Data Mining

Data mining is the science and technology of exploring large and complex bodies of data in order to discover useful patterns. It is extremely important because it enables modeling and knowledge extraction from abundant data availability. Soft Computing for Knowledge Discovery and Data Mining introduces soft computing methods extending the envelope of problems that data mining can solve efficiently. It presents practical soft-computing approaches in data mining. This edited volume by highly regarded authors, includes several contributors of the 2005, Data Mining and Knowledge Discovery Handbook.This bookwas written to provide investigators in the fields of information systems, engineering, computer science, statistics and management with a profound source for the role of soft computing in data mining. Not only does this book feature illustrations of various applications including manufacturing, medical, banking, insurance and others, but also includes various real-world case studies with detailed results. Soft Computing for Knowledge Discovery and Data Mining is designed for practitioners and researchers in industry. Practitioners and researchers may be particularly interested in the description of real world data mining projects performed with soft computing. This book is also suitable as a secondary textbook or reference for advanced-level students in information systems, engineering, computer science and statistics management.

Decomposition methodology for knowledge discovery and data mining : theory and applications

Introduction to Data Mining Decision Trees Clustering Techniques Ensemble Methods Decomposition Methodology in Data Mining Feature Set Decomposition Space Decomposition Sample Decomposition Function Decomposition Concept Decomposition Automatic Decomposition Conclusions, Advanced Issues and Open Questions.