International Journal on Artificial Intelligence Tools

Standard Electrocardiogram (ECG) database is created for validating and comparing different algorithms on feature detection and disease classification. At present, there are four frequently used standard databases: MIT-BIH arrhythmia database, QT database, CSE multi-lead database and AHA database. With the development in equipment and diagnosis approach, severe deficiencies are discovered and a new modern ECG database is needed for further research. So Chinese Cardiovascular Disease Database (CCDD or CCD database), which contains 12-Lead ECG data, detailed annotation features and beat diagnosis result is proposed. It is advanced for not only improving the raw ECG data's technical parameters, but also introducing valuable morphology features which are utilized by experienced cardiologists effectively. CCDD is employed by our group as well as aiming for supporting other research groups that work in automated ECG analysis.

Over the past decade Wireless Capsule Endoscopy (WCE) technology has become a very useful tool for diagnosing diseases within the human digestive tract. Using WCE physicians can examine the digestive tract in a minimum invasive way searching for pathological abnormalities such as bleeding, polyps, ulcers and Crohn's disease. In order for WCE to be more effective for gastroenterologists, engineers have developed software methods to automatically detect these diseases at high successful rate. Using proposed a synergistic methodology for automatic discovering polyps (protrusions) and ulcers in WCE video frames, a data mining approach is used that offers useful information about ulcers, polyps and normal tissues and their visual similarities. Finally, results of the methodology are given and statistical comparisons are also presented relevant to other works.

The early identification of mild cognitive impairment (MCI) has the potential for timely therapeutic interventions that would limit the advancement of MCI to Alzheimer's disease (AD). This paper presents an automated approach for early detection of MCI through pattern classification of magnetic resonance imaging (MRI) data. The approach is based on image feature selection and support vector machine (SVM) classification. Subjects were selected from the Open Access Series of Imaging Studies (OASIS) database and included 89 MCI subjects and 80 controls. Voxel-by-voxel differences in gray matter (GM) intensity between the MCI and control groups were identified. Then regions of interest (ROIs) and the most discriminative image features that represented the patterns in MCI subjects were determined for training a classifier. The classifier demonstrated a high classification accuracy (90%) when a behavioral estimate of MCI and the ROIs were included as features in comparison to the behavioral estimate or the ROIs alone, which is one scientific contribution of our work. Another contribution is that the classifier can be integrated with the image processing functions through an online interface with significant medical capability that can be used for automated image pre-processing, obtaining MCI probability estimates for individual cases, and visualization of affected regions.

The prediction of gene function from genome sequences is one of the main issues in Bioinformatics. Most computational approaches are based on the similarity between sequences to infer gene function. However, the availability of several fully sequenced genomes has enabled alternative approaches, such as phylogenetic profiles. Phylogenetic profiles are vectors which indicate the presence or absence of a gene in other genomes. The main concept of phylogenetic profiles is that proteins participating in a common structural complex or metabolic pathway are likely to evolve in a correlated fashion. In this paper, a multi-level clustering algorithm of phylogenetic profiles is presented, which aims to detect inter- and intra-genome gene clusters.

This paper investigates and compares the performance of local descriptors for race classification from face images. Two powerful types of local descriptors have been considered in this study: Local Binary Patterns (LBP) and Weber Local Descriptors (WLD). First, we investigate the performance of LBP and WLD separately and experiment with different parameter values to optimize race classification. Second, we apply the Kruskal-Wallis feature selection algorithm to select a subset of more "discriminative" bins from the LBP and WLD histograms. Finally, we fuse LBP and WLD, both at the feature and score levels, to further improve race classification accuracy. For classification, we have considered the minimum distance classifier and experimented with three distance measures: City-block, Euclidean, and Chi-square. We have performed extensive experiments and comparisons using five race groups from the FERET database. Our experimental results indicate that (i) using the Kruskal-Wallis feature selection, (ii) fusing LBP with WLD at the feature level, and (iii) using the City-block distance for classification, outperforms LBP and WLD alone as well as methods based on holistic features such as Principal Component Analysis (PCA) and LBP or WLD (i.e., applied globally).

Speech recognition is an efficient tool for interacting between human and machines; but the speech includes some undesirable components (disfluencies), sometimes, and is called spontaneous speech in these cases. This paper focuses on detecting some simple disfluencies in this speech type, using a Learning Cellular Automata. This is a set of cells that each is a learning automata (LA), and selects an action with a determined probability. Each LA is assigned to a word of the speech transcription sentence, and the action set for each LA includes four simple disfluencies types and "Fluent". The environment considered for interacting with cells, evaluates the action selected by each cell in every step, and sends back a response to the LA. That response will be considered for rewarding or penalizing the selected type. Because of parallel operation by the cells, and by selecting appropriate values for rewarding and penalizing coefficients, the proposed model can reach a desirable performance for tagging the disfluencies, regarding the time consumption and precision.

Many decision tree algorithms were proposed over the last few decades. A lack of publishing standards for decision tree algorithm software produced a large time gap between algorithm proposals and their wider application in practice. Non-existence of common repository for storing algorithms and their parts led to a need to re-implement these algorithms from a scratch when they had to be implemented on a different platform. This makes the comparison between algorithms and their partial improvements vague. In addition, combinations and interactions between different algorithm parts haven't been analyzed thoroughly. Reusable component design of decision tree algorithms has been recently suggested as a potential solution to these problems. In this paper we describe an architecture for component-based (white-box) decision tree algorithm design, and we present an open-source framework which enables design and fair testing of decision tree algorithms and their parts. This architecture and developed platform can provide the research community with a common codebase for storing, designing, and evaluating decision tree algorithms (traditional, multivariate and hybrid) and their partial improvements. It is intended for data mining practitioners, algorithm and software developers, and as well for students, as a technology enhanced learning tool.

Decision-making on uncertain and dynamic domains is still a challenging research area. This paper explores a solution to handle such complex decision making based on a combined logic system. We provide an explanation of our reasoning system focused on the algorithms and their implementations. The reasoning system is based on a multi-valued temporal propositional logic which we use as the foundation for the implementation of simulation/prediction and query answering tools. This system is available for users to represent knowledge and to refine these systems to debug them and to try different problem solving strategies. We provide examples to illustrate how the system can be used including a problem based on a real smart environment.

The present paper describes a Bayesian network approach to Information Retrieval (IR) from Web documents. The network structure provides an intuitive representation of uncertainty relationships and the embedded conditional probability table is used by inference algorithms in an attempt to identify documents that are relevant to the user's needs, expressed in the form of Boolean queries. Our research has been directed in constructing a probabilistic IR framework that focus on assisting users to perform Ad-hoc retrieval of documents from the various domains such as economics, news, sports, etc. Furthermore, users can integrate feedback regarding the relevance of the retrieved documents in an attempt to improve performance on upcoming requests. Towards these goals, we have expanded the traditional Bayesian network IR system and tested it on several Greek web corpora on different application domains. We have developed two different approaches with regards to the structure: a simple one, where the structure is manually provided, and an automated one, where data mining is used in order to extract the network's structure. Results have depicted competitive performance against successful IR models of different theoretical backgrounds, such as the vector space utilizing tf-idf and the probabilistic model of BM25 in terms of precision-recall curves. In order to further improve the performance of the IR system, we have implemented a novel similarity-based lemmatization framework, reducing thus the ambiguity posed by the plethora of morphological variations of the languages in question. The employed lemmatization framework comprises of 3 core components (i.e. the word segregation, the data cleansing and the lemmatization modules) and is language-independent (i.e. can be applied to other languages with morphological peculiarities and thus improve Ad-hoc retrieval) since it achieves the mapping of an input word to its normalized form by employing two state-of-the-art language independent distance metric models, meaning the Levenshtein Edit distance and the Dice coefficient similarity measure, combined with a language model describing the most frequent inflectional suffixes of the examined language. Experimental results support our claim on the significance of this incorporation to Greek texts web retrieval as results improve by a factor of 4% to 11%.