DOI: 10.3897/jucs.175197

Authors: Sooin Kim, Kyungtae Kim, Donghoon Kim, Doosung Hwang

Abstract: Conventional classification methods rely on the assumption that samples are independent and identically distributed, often ignoring the latent relational structures governing real-world data. Consequently, these methods fail to capture cross-sample dependencies, resulting in impoverished feature representations and suboptimal generalization. To address this limitation, we propose the Feature-refinement Classification Head (FCH), a novel and modular component that explicitly models inter-sample relationships within training batches. FCH constructs an adjacency matrix from embedding vectors and employs a graph neural network to refine features via relational propagation. Seamlessly integrated and architecture-agnostic, FCH leverages the end-to-end optimization using cross-entropy and structure preserving loss as joint objectives. Extensive experiments on benchmarks such as CIFAR-10, MNIST, and STL-10 demonstrate that FCH consistently enhances accuracy, precision, recall, and F1-scores across diverse backbones. Notably, FCH demonstrates significant performance improvements even in lightweight models, demonstrating its robustness, scalability, and practicality. Overall, FCH offers a principled, generalizable alternative to traditional classification heads by effectively leveraging sample-to-sample relationships.

HTML

PDF

DOI: 10.3897/jucs.161202

Authors: Yelda Fırat, Yılmaz Kılıçaslan, Hüseyin Ali Sarıkaya, Murat Kaan Yılmaz

Abstract: Autism Spectrum Disorder is a neurodevelopmental disorder with onset in early childhood and its diagnosis often requires clinical processes based on long, subjective observations. Although early diagnosis and intervention can significantly improve developmental outcomes, existing methods are limited in terms of scalability and objectivity. The aim of this study is to develop a hybrid deep learning model that detects Autism Spectrum Disorder with high accuracy by analyzing motor behaviors from videos of children recorded in their natural home environment. In this study, joint coordinates were extracted using the MediaPipe Pose model and spatial, temporal, frequency and coordination-based features were calculated from these data. The features were processed with a hybrid architecture integrating CNN, BiLSTM and attention mechanism. CNN captured spatial patterns, BiLSTM learned the dynamics over time, and the attention mechanism focused on critical movement segments. The model achieves over 97% accuracy on closed datasets and over 83% on public videos such as YouTube and TikTok. These results show that the method performs robustly under both controlled and real-world conditions. The study provides a scalable, objective and clinically applicable screening tool that overcomes the problems of artificial environments and limited data.

HTML

PDF

DOI: 10.3897/jucs.161636

Authors: Wicus J. van der Linden, Trienko L. Grobler, Lynette van Zijl

Abstract: The effect of imbalanced data on the optical character recognition of Braille text is investigated by applying two techniques to a set of convolutional neural network image classification models. A multilabel classification framework is applied to identify the combination of Braille dots present in a character sample. This approach is compared to the multiclass classification framework prevalent in the literature, which directly identifies each sample as one of 64 possible Braille characters. Furthermore, data resampling methods are applied to investigate the impact of class imbalance on the multilabel and multiclass modelling approaches, respectively. The multilabel models are shown to achieve statistically significantly better performance than multiclass models, across different data resampling strategies. This includes better generalisation to out of distribution testing data from different Braille language codes, as well as robust performance under experimental image augmentation conditions. Furthermore, while multiclass models achieve better performance when trained on resampled data compared to training without resampling, this performance increase fails to rival the performance of the multilabel classification models across metrics and resampling strategies.

HTML

PDF

DOI: 10.3897/jucs.150393

Authors: Alexandru Pintea

Abstract: Smart homes, live streaming IoT devices, and smart sensors can all be optimised to enhance energy efficiency. In order to offer a cheap alternative to the traditional real-time monitoring systems, this study proposes a sensor-based occupancy system. The evaluation in real time of the number of occupants in buildings/ rooms /houses is reflected in the energy usage. Sensor data can provide insight into many characteristics of a considered environment. The sensor dataset considered was collected with the aim of determining how many people are present in a given space/room. The sensor data does not portray the people present in the room, but rather their impact on it (e.g. CO2/ noise/ light level changes). The dataset was cleaned and preprocessed to optimise model performance. The results obtained by training several classifiers yielded accuracies that reach 98%-99%. The research provides an end-to-end solution for the considered problem, through data preprocessing/feature selection/outlier removal and model training/evaluation. Hyperparameters were tuned for more than twenty models. All chosen models and features were ranked based on performance and robustness. A novel solution for optimising sensor placement has also been proposed by this study, to further improve sensor-based monitoring systems.

HTML

PDF

DOI: 10.3897/jucs.130450

Authors: Jafseer KT, Shailesh S, Sreekumar A

Abstract: Recent advancements in data mining and knowledge discovery have created numerous research opportunities in streaming data analysis. One critical challenge is developing machine learning models that can efficiently handle changes in features and dynamic concepts, including concept drift, feature drift, and feature evolution. State-of-the-art techniques proposed to address these anomalies in data streams often assume that a constant set of features is available for processing. However, in real-time scenarios, the situation is quite different, as the set of features in a stream may vary over time due to factors such as the disappearance of existing features or the emergence of new ones. The proposed work focuses on handling dynamically evolving features by introducing a novel solution that leverages a Dynamic Autoencoder DAE and ensemble learning. Additionally, adaptive windowing and concept-preserving mechanisms improve the proposed architecture by retaining the concept information from previous data windows. The ensemble technique used in the proposed classification framework demonstrates promising performance in diverse datasets, achieving accuracies of 86%, 94%, and 95% in the Weather, Electricity and Forest Cover Type datasets, respectively. This innovative integration of deep learning and traditional methods effectively addresses various challenges in streaming data analysis.

HTML

PDF

DOI: 10.3897/jucs.137544

Authors: Igor Tomičić, Petra Grd, Andrija Bernik

Abstract: This paper presents a comprehensive analysis of the integration of artificial intelligence (AI) into threat intelligence (TI) systems, focusing on its potential to enhance cybersecurity operations. The paper presents an extensive literature review, covering the current state of AI applications in TI, including machine learning, deep learning, and natural language processing for automating threat detection, classification, and analysis. It outlines the core functions of AI in TI, such as threat detection, correlation, prediction, and automated response, and introduces a detailed taxonomy categorizing AI techniques based on their roles in enhancing TI processes. Additionally, the paper proposes a conceptual workflow for AI-powered TI, illustrating how AI can streamline data collection, threat analysis, and incident response. A roadmap for future research is further provided, highlighting key areas for development, including explainable AI, federated learning, and edge computing. This work contributes to the field by offering a structured framework for integrating AI into TI and identifying critical challenges and future directions in AI-driven cybersecurity.

HTML

PDF

DOI: 10.3897/jucs.114357

Authors: Samit Bhanja, Banani Ghose, Abhishek Das

Abstract: Recently time series forecasting has become one of the prime application areas of climatology, economics and industries. Many research works are conducted to forecast the time series more accurately. But few of them are concentrated on predicting the time series over an extended future horizon, and there is also a scope to improve their forecasting accuracy. This work proposes a multi-step-ahead foresting method to produce a stable and accurate forecasting result for the extended future horizon. Firstly, a deep learning-based forecasting model is proposed to predict the time series. Secondly, a fuzzy time series-based error correction model is implemented to enhance the prediction performance of the deep learning model. Here to optimize all the fuzzy time series (FTS) parameters in an integrated way, an integrated butterfly optimization (FTS-IBO) algorithm is proposed. In this study, two different types of real-world multivariate time series datasets are used to analyze the forecasting performance of the proposed model. The performance of the proposed FTS-IBO algorithm is compared with the traditional butterfly optimization (FTS-BO) algorithm. The experimental results show that the FTS-IBO technique is superior to the FTS-BO technique. The forecasting performance of the proposed model has also compared the other state-of-the-art models, and the simulation results exhibit that the proposed model produces a more accurate prediction performance for multi-step-ahead time series forecasting problems compared to other models.

HTML

PDF

DOI: 10.3897/jucs.119234

Authors: Panji Bintoro, Zulkifli Zulkifli, Yaya Heryadi, Fitriana Fitriana, Nopi Anggista Putri, Dwi Yana Ayu Andini

Abstract: Mpox is a disease like smallpox caused by the Mpox virus (MPXV), which belongs to the Orthopoxvirus (OPXV) group in the Poxviridae family. The virus is transmitted through direct contact with infected individuals, animals, or contaminated materials. Transmission can occur through direct body contact, animal bites, respiratory droplets, or mucous membranes in the eyes, nose, or mouth. However, since the recent outbreak in May 2022, the disease has spread to various countries, posing a threat to develop into a global pandemic. Several image processing and deep learning models, including Convolutional Neural Network (CNN), have been employed for Mpox disease prediction. The default CNN algorithm performs poorly on image orientations such as tilting, rotation, zooming, or other abnormal images. Therefore, we propose a new framework adopted from deep learning by combining Generative Adversarial Network (GAN), PyramidalNet, and Long Short-Term Memory (LSTM). This new method is referred to as GPLNet. The research results indicate that the GPLNet algorithm model can surpass the accuracy achieved by CNN and CNN-LSTM, reaching 99%. The performance of the GPLNet algorithm model is also evaluated using various measurement metrics, yielding an accuracy of 98%, precision of 99%, recall of 98%, sensitivity of 98%, specificity of 98%, f1-score of 98%, and ROC of 99%.

HTML

PDF

DOI: 10.3897/jucs.109227

Authors: Binnu Paul, Dwijen Rudrapal, Kunal Chakma, Anupam Jamatia

Abstract: Multimodal machine translation (MMT) is a challenging task in the linguistically diverse Indian landscape. Machine translation refers to the task of automatically converting content from one language to another without human involvement. Within the realm of natural language processing, a significant challenge arises from the inherent ambiguity present in human language. Translation ambiguity is a cross-lingual phenomenon that can manifest itself for various reasons, including lexical ambiguity, the occasional need to impute missing words, the presence of gen-der ambiguity, and word-sense ambiguities. These factors can lead to a decrease in translation accuracy. The integration of multiple modalities, such as images, videos, and audio, in addition to text, plays a pivotal role in improving the robustness and precision of translation systems. Over the past five years, extensive research has been dedicated to incorporating secondary modalities alongside text to improve language translation and comprehension. In this comprehensive study, our objective was to identify and explore promising MMT approaches, available corpora, eval-uation metrics, research challenges, and the future direction of research specifically for Indian languages. We evaluated 81 papers, including MMT models, MMT dataset in Indian languages, survey on MMT approach, and the effects of multiple modalities in machine translation. The performance of the different proposed approaches has also been briefly analyzed on the basis of the claimed results and comparative evaluations. Finally, the challenges associated with the MMT task for India and some possible directions for future research in this domain are highlighted.

HTML

PDF

DOI: 10.3897/jucs.107081

Authors: Alaor Cervati Neto, Alexandre L. M. Levada

Abstract: Dimensionality reduction based unsupervised metric learning consists in finding meaningful compact data representations previously to clustering and classification problems. One of the major aspects of these algorithms is the approximation of the underlying manifold by a weighted graph. A limitation with most manifold learning algorithms is that edge weights in the proximity graph rely heavily on the Euclidean distance, which is known to be quite sensitive to the presence of outliers. In this paper, we propose to improve the Locality Preserving Projections (LPP) algorithm by incorporating a recently proposed graph inference method called Probabilistic Nearest Neighbors (PNN), an extension of the Clustering with Adaptive Neighbors (CAN) approach, used with success in graph-based semi-supervised learning. The proposed PNN-LPP algorithm is able to achieve better classification results than regular LPP, showing competitive performance against state-of-the-art approaches for dimensionality reduction, such as the UMAP algorithm, especially in datasets with a limited number of samples.

HTML

PDF

DOI: 10.3897/jucs.89923

Authors: Stefan Strydom, Andrei Michael Dreyer, Brink van der Merwe

Abstract: International Classification of Disease (ICD) coding plays a significant role in classify-ing morbidity and mortality rates. Currently, ICD codes are assigned to a patient’s medical record by hand by medical practitioners or specialist clinical coders. This practice is prone to errors, and training skilled clinical coders requires time and human resources. Automatic prediction of ICD codes can help alleviate this burden. In this paper, we propose a transformer-based architecture with label-wise attention for predicting ICD codes on a medical dataset. The transformer model is first pre-trained from scratch on a medical dataset. Once this is done, the pre-trained model is used to generate representations of the tokens in the clinical documents, which are fed into the label-wise attention layer. Finally, the outputs from the label-wise attention layer are fed into a feed-forward neural network to predict appropriate ICD codes for the input document. We evaluate our model using hospital discharge summaries and their corresponding ICD-9 codes from the MIMIC-III dataset. Our experimental results show that our transformer model outperforms all previous models in terms of micro-F1 for the full label set from the MIMIC-III dataset. This is also the first successful application of a pre-trained transformer architecture to the auto-coding problem on the full MIMIC-III dataset.

HTML

XML

PDF

DOI: 10.3897/jucs.2020.037

Authors: Oscar Camacho-Nieto, Cornelio Yáñez-Márquez, Yenny Villuendas-Rey

Abstract: This paper proposes a novel undersampling method, for dealing with imbalanced datasets. The proposal is based on a novel instance importance measure (also introduced in this paper), and is able to balance hybrid and incomplete data. The numerical experiments carried out show the proposed undersampling algorithm outperforms others algorithms of the state of art, in well-known imbalanced datasets.

HTML

XML

PDF

DOI: 10.3897/jucs.2020.035

Authors: Indraja Germanaitė, Kętutis Zaleckis, Rimantas Butleris, Kristina Jarmalavičienė

Abstract: In this case study the authors created and tested a configurable and expandable spatial patterns (SP) description, identification, and application methodology (SPDIAM) and an SP identification algorithm. SPDIAM allows urban planning and design (UPD) practitioners to describe SP in a computerized manner, identify SP automatically and then apply them in the UPD domain. SPDIAM is based on the space syntax (SS) method and normalized spatial and non-spatial measures and can be used with the statistical social, economic, and environmental indicators, which are related to the urban sustainability and spatial capital. The goal of the case study experiment was to proof a concept of SPDIAM and to identify the rules and the values of the measures used for the SP identification. For this City Layout SP was identified in the vector data of 12 European, North American, and African cities. The experiment results confirmed that SPDIAM is appropriate to describe SP and identify them automatically. The use of the normalized measures enables the comparison of different SP and reduces the degree of the subjectivity of the UPD solutions. SPDIAM no longer relies on statistical information but forms SP based on the probabilistic complex modelling of a city, which lets SPDIAM indicate possible directions of SP future transformation. SPDIAM uses the newly offered measures CENTER and URBAN COMPACTNESS INDEX to identify SP automatically and can add quantitative and qualitative improvement to the spatial network analysis tools in Geographic Information Systems.

HTML

XML

PDF

DOI: 10.3217/jucs-025-06-0667

Authors: Sonia Contreras, Miguel Manzanedo, Álvaro Herrero

Abstract: Workplace accident rates are always a constant source of concern in different scenarios. On the other hand, different economic cycles modify employment indicators, thereby affecting both the health and the wellbeing of workers. In this study, a Hybrid Neural System is proposed to support the analysis of both workplace accidents and different macroeconomic variables. The application of exploratory projection and nonlinear autoregressive models allow us to recognize patterns and subsequently study the interplay between the two fields, providing valuable information, so that directors can take more informed management decisions.

HTML

XML

PDF

DOI: 10.3217/jucs-024-10-1378

Authors: Francois Meyer, Brink van der Merwe, Dirko Coetsee

Abstract: Word embedding techniques can be used to learn vector representations of concepts from temporal datasets. Previous attempts to do this amounted to appling word embedding techniques to event sequences. We propose a concept embedding model that extends existing word embedding techniques to take time into account by explicitly modelling the time between concept occurrences. The model is implemented and evaluated using medical temporal data. It is found that incorporating time into the learning algorithm can improve the quality of the resulting embeddings, as measured by an existing methodological framework for evaluating medical concept embeddings.

HTML

XML

PDF

DOI: 10.3217/jucs-024-06-0775

Authors: Yenny Villuendas-Rey, Mariana-D. Alanis-Tamez, Carmen-F. Benguría, Cornelio Yáñez-Márquez, Oscar Camacho-Nieto

Abstract: Computational Intelligence techniques in medicine have become an increasing area of research worldwide. Among them, the application and development of new models and algorithms for disease diagnosis and prediction have been an active research topic. The research contribution of the current paper is the proposal of a novel classification model, and its application to the diagnosis of chronic diseases. One of the main characteristics of the new model is that it is designed to deal with imbalanced data. With the purpose of making experimental comparisons to demonstrate the benefits of the proposed model, we tested five classification models, over medical data. The application of the supervised classification algorithms is done over the Knowledge Extraction based on Evolutionary Learning (KEEL) environment, using a distributed optimally balanced stratified 5-fold cross validation scheme. In addition, the experimental results obtained were validated in order to identify significant differences in performance by mean of a non-parametric statistical test (the Friedman test), and a post-hoc test (the Holm test). The hypothesis testing analysis of the experimental results indicates that the proposed model outperforms other supervised classifiers for medical diagnosis.

HTML

XML

PDF

DOI: 10.3217/jucs-024-04-0460

Authors: Dawid Połap

Abstract: The increasing need to implement systems as identity verification meant that safe and effective verification systems have become one of the main branches of modern scientiffic research. In this paper, the idea of identity verification support system is presented. The main novelty of this work is to present the technique of feature extraction that can be implemented simultaneously for processed image files as well sound ones. Proposed method was tested due to achieve the highest possible accuracy of the selected neural classifier. The obtained results were discussed in terms of advantages and disadvantages for practical applications.

HTML

XML

PDF

DOI: 10.3217/jucs-023-03-0304

Authors: Sonia Ortiz-Ángeles, Yenny Villuendas-Rey, Itzamá López-Yáñez, Oscar Camacho-Nieto, Cornelio Yáñez-Márquez

Abstract: The contemporary world has witnessed technological advances, such as Online Social Networks (OSN), whose influence in almost every action of the human being is remarkable. Among the human activities most significantly impacted by OSNs are: entertainment, human relationships, education, and political activities, including those related to electoral campaigns and electoral preferences prediction. The research contribution of the current paper regards the usefulness of OSNs users generated data to predict the political context. More specifically, 25 Computational Intelligence (CI) algorithms are used to predict voting intentions on the United States primary presidential elections for 2016, taking as input the data sets generated by 1200 users of the YouGov OSN, as well as the answers they gave to an online study run by the American National Election Studies (ANES). The application of the 25 supervised classification algorithms is done over the Waikato Environment for Knowledge Analysis (WEKA), using a stratified 5-fold cross validation scheme. Also, the experimental results obtained were validated in order to identify significant differences in performance by mean of a non-parametric statistical test (the Friedman test), and a post-hoc test (the Holm test). The hypothesis testing analysis of the experimental results indicates that predicting voting intentions in favour of a democrat or republican candidate is simpler than predicting the particular candidate, given that the prediction performances for a democrat or republican candidate (best performances of 80% and 78%, respectively) are better than those given when predicting a specific candidate (70% for democrat candidates and 56% for republican candidates).

HTML

XML

PDF

DOI: 10.3217/jucs-019-15-2224

Authors: Wedad Hussein, Rasha Ismail, Tarek Gharib, Mostafa G. M. Mostafa

Abstract: Recommender systems are means for web personalization and tailoring the browsing experience to the users' specific needs. There are two categories of recommender systems; memory-based and model-based systems. In this paper we propose a personalized recommender system for the next page prediction that is based on a hybrid model from both categories. The generalized patterns generated by a model based techniques are tailored to specific users by integrating user profiles generated from the traditional memory-based system's user-item matrix. The suggested system offered a significant improvement in prediction speed over traditional model-based usage mining systems, while also offering an average improvement in the system accuracy and system precision by 0.27% and 2.35%, respectively.

HTML

XML

PDF

DOI: 10.3217/jucs-019-03-0325

Authors: Pekka Kekolahti, Juuso Karikoski

Abstract: The purpose of this paper is to identify probabilistic relationships of mobile service usage behaviour, and especially to understand the probabilistic relationship between overall service usage diversity and average daily service usage intensity. These are topical themes due to the high number of services available in application stores which may or may not lead to high usage diversity of mobile services. Four analytical methods are used in the study, all are based on Bayesian Networks; 1) Visual analysis of Bayesian Networks to find initially interesting patterns, variables and their relationships, 2) user segmentation analysis, 3) node force analysis and 4) a combination of expert-based service clustering and machine learning for usage diversity vs. intensity analysis. All the analyses were conducted with handset-based data collected from university students and staff. The analysis indicates that services exist, which mediate usage of other services. In other words, usage of these services increases the probability of using also other services. A service called Installer is an example of this kind of a service. In addition, probabilistic relationships can be found within certain service cluster pairs in their usage diversity and intensity values. Based on these relationships, similar mediation type of behaviour can be found for service clusters as for individual services. This is most visible in the relation between System/Utilities and Business/Productivity service clusters. They do not have a direct relationship but usage diversity is a mediator between them. Furthermore, segmentation analysis shows that the user segment called "experimentalists" uses more mediator services than other user segments. Furthermore, "experimentalists" use a much broader set of services daily, than the other segments. This study demonstrates that a Bayesian Network is a straightforward way to express model characteristics on high level. Moreover, Node Force, Direct and Total effect are useful metrics to measure the mediation effects. The clustering implemented as a hybrid of machine learning and expert-based clustering process is also a useful way to calculate relationships between clusters of more than a hundred individual services.

HTML

XML

PDF

DOI: 10.3217/jucs-018-04-0577

Authors: Anna Zygmunt, Piotr Bródka, Przemysław Kazienko, Jarosław Koźlak

Abstract: Identifying key persons active in social groups in the blogosphere is performed by means of social network analysis. Two main independent approaches are considered in the paper: (i) discovery of the most important individuals in persistent social communities and (ii) regular centrality measures applied either to social groups or the entire network. A new method for separating of groups stable over time, fulfilling given conditions of activity level of their members is proposed. Furthermore, a new concept for extracting user roles and key persons in such groups is also presented. This new approach was compared to the typical clustering method and the structural node position measure applied to rank users. The experimental studies have been carried out on real two-year blogosphere data.

HTML

XML

PDF

DOI: 10.3217/jucs-018-04-0507

Authors: Onur Can Sert, Kayhan Dursun, Tansel Özyer, Jamal Jida, Reda Alhajj

Abstract: Multi objective clustering is one focused area of multi objective optimization. Multi objective optimization attracted many researchers in several areas over a decade. Utilizing multi objective clustering mainly considers multiple objectives simultaneously and results with several natural clustering solutions. Obtained result set suggests different point of views for solving the clustering problem. This paper assumes all potential solutions belong to different experts and in overall; ensemble of solutions finally has been utilized for finding the final natural clustering. We have tested on categorical datasets and compared them against single objective clustering result in terms of purity and distance measure of k-modes clustering. Our clustering results have been assessed to find the most natural clustering. Our results get hold of existing classes decided by human experts.

HTML

XML

PDF

DOI: 10.3217/jucs-017-06-0944

Authors: Sang Lee, Dong Kim, Jong Park

Abstract: E-mail spam is no more garbage but risk since it recently includes virus attachments and spyware agents which make the recipients' system ruined, therefore, there is an emerging need for spam detection. Many spam detection techniques based on machine learning techniques have been proposed. As the amount of spam has been increased tremendously using bulk mailing tools, spam detection techniques should counteract with it. To cope with this, parameters optimization and feature selection have been used to reduce processing overheads while guaranteeing high detection rates. However, previous approaches have not taken into account feature variable importance and optimal number of features. Moreover, to the best of our knowledge, there is no approach which uses both parameters optimization and feature selection together for spam detection. In this paper, we propose a spam detection model enabling both parameters optimization and optimal feature selection; we optimize two parameters of detection models using Random Forests (RF) so as to maximize the detection rates. We provide the variable importance of each feature so that it is easy to eliminate the irrelevant features. Furthermore, we decide an optimal number of selected features using two methods; (i) only one parameters optimization during overall feature selection and (ii) parameters optimization in every feature elimination phase. Finally, we evaluate our spam detection model with cost-sensitive measures to avoid misclassification of legitimate messages, since the cost of classifying a legitimate message as a spam far outweighs the cost of classifying a spam as a legitimate message. We perform experiments on Spambase dataset and show the feasibility of our approaches.

HTML

XML

PDF

DOI: 10.3217/jucs-017-06-0926

Authors: Marcin Piekarczyk, Marek Ogiela

Abstract: One important subject associated with personal authentication capabilities is the analysis of handwritten signatures. Among the many known techniques, algorithms based on linguistic formalisms are also possible. However, such techniques require a number of algorithms for intelligent image analysis to be applied, allowing the development of new solutions in the field of personal authentication and building modern security systems based on the advanced recognition of such patterns. The article presents the approach based on the usage of syntactic methods for the static analysis of handwritten signatures. The graph linguistic formalisms applied, such as the IE graph and ETPL(k) grammar, are characterised by considerable descriptive strength and a polynomial membership problem of the syntactic analysis. For the purposes of representing the analysed handwritten signatures, new hierarchical (two-layer) HIE graph structures based on IE graphs have been defined. The two-layer graph description makes it possible to take into consideration both local and global features of the signature. The usage of attributed graphs enables the storage of additional semantic information describing the properties of individual signature strokes. The verification and recognition of a signature consists in analysing the affiliation of its graph description to the language describing the specimen database. Initial assessments display a precision of the method at a average level of under 75%.

HTML

XML

PDF

DOI: 10.3217/jucs-015-04-0705

Authors: Jeen-Shing Wang, Jen-Chieh Chiang

Abstract: This paper presents an efficient data preprocessing procedure for the of support vector clustering (SVC) to reduce the size of a training dataset. Solving the optimization problem and labeling the data points with cluster labels are time-consuming in the SVC training procedure. This makes using SVC to process large datasets inefficient. We proposed a data preprocessing procedure to solve the problem. The procedure contains a shared nearest neighbor (SNN) algorithm, and utilizes the concept of unit vectors for eliminating insignificant data points from the dataset. Computer simulations have been conducted on artificial and benchmark datasets to demonstrate the effectiveness of the proposed method.

HTML

XML

PDF

DOI: 10.3217/jucs-014-15-2582

Authors: Yudel Gómez, Rafael Bello, Amilkar Puris, María García, Ann Nowe

Abstract: In this paper we propose a new approach to Swarm Intelligence called Two-Step Swarm Intelligence. The basic idea is to split the heuristic search performed by agents into two stages. In the first step the agents build partial solutions which, are used as initial states in the second step. We have studied the performance of this new approach for the Feature Selection Problem by using Ant Colony Optimization and Particle Swarm Optimization. The feature selection is based on the reduct concept of the Rough Set Theory. Experimental results obtained show that Two-step approach improves the performance of ACO and PSO metaheuristics when calculating reducts in terms of computation time cost and the quality of reducts.

HTML

XML

PDF

DOI: 10.3217/jucs-004-07-0630

Authors: Alberto Faro, Daniela Giordano, Corrado Santorno

Abstract: The paper proposes a neural agent that performs self-organizing classification to assist in searching and contributing to webs of documents and in the process of document reuse. By applying the Kohonen self-organizing feature map (SOFM) algorithm to patterns of influence links among documents it is possible to originate clusters of documents that help infer the aspects that such documents implicitly share. The approach complements search techniques based on semantic indexes. The resulting classification is sensitive to the multiple aspects of a document that may belong to multiple classes with a varying degree and allows for treating effectively items that typically have a limited life span, either because they are means to the collaborative production of a more complex item, or because they belong to fast evolving domains. The method has been implemented by Lotus Notes Domino Web server for a case-based application in the domain of information systems design.

HTML

XML

PDF