Letteratura scientifica selezionata sul tema "MACHINE LARNING"

Crdit risk plays an important rol in financ. In a sns, crdit risk rflcts th stability of financial institutions and th protction of invstors. Du to th impact of th pidmic on th world conomy, w nd to rassss th crdit risk. For a long tim, machin larning and dp larning in statistics hav bn vry ffctiv in prdicting crdit risk. But in machin larning and dp larning, whn prdicting crdit risk, w citd four modls, namly XGBoost, Dcision Tr, Random Forst and Convolutional Nural Ntwork (CNN) modls. Analyz th advantags and disadvantags of ths modls: XGBoost can optimiz th tr modl and prvnt ovrfitting through rgularization, but XGBoost has poor intrprtability. Dcision Tr has strong xplanatory powr, but it is prone to overfitting. Compared with th dcision tr, Random Forsts incras accuracy and rduc th probability of overfitting, but Random Forsts consum mor tim and computing rsourcs. Although Convolutional Nural Ntwork has a high accuracy rat, it abandons intrprtability. Thrfor, in our xprimnts, w found that ths modls ar not prfct and hav thir own dfcts. So, in futur rsarch, w will mak an intgratd modl to includ th advantags of th modl and discard som dfcts, so that th intgratd modl will hav bttr gnralization and accuracy.

Liver merupakan organ tubuh manusia yang memiliki peranan sangat penting seperti mencerna, menyerap, membantu proses pencernaan makanan serta menghancurkan racun di dalam darah. Penyakit hati atau liver yang sudah akut sangat mempengaruhi fungsi-fungsi hati, penyakit hati dapat diketahui dari munculnya gejala klinis maupun fisik yang timbul pada pasien. Penelitian ini membahas tentang klasifikasi penyakit liver pada dataset ILPD yang diambil dari UCI Machine learning Repository menggunakan algoritma machine learning. Dataset terdiri dari 583 record data, 10 kriteria, dan 1 variable kelas berjenis multivariate. Penelitian ini menggunakan beberapa tahapan preprocessing yang dilakukan, diantaranya : Preprocessing Data Dan Eksplorasi Data, Penanganan missing value, feature selection, menerapkan feature correlation dan feature scaling, Analisis menggunakan Algoritma Machine learning. Berdasarkan hasil pengujian yang dilakukan dalam memperoleh nilai akurasi perhitungan klasifikasi menggunakan Algoritma Random Forest memiliki performa keakuratan yang diukur dengan akurasi sebesar 78,63% sehingga disimpulkan akurasi tersebut lebih unggul dari algoritma lainnya dalam klasifikasi penyakit liver.

A pesquisa aborda o uso e os estudos sobre Inteligência Artificial (IA) voltada para as tarefas da Ciência da Informação (CI), seu objetivo é analisar os últimos 20 anos da produção científica brasileira a respeito da IA, os objetivos específicos são identificar as abordagens da IA, da CI, os autores, as instituições, as revistas cientificas, as áreas profissionais, o volume de publicações e as metodologias empregadas nos estudos por meio de uma análise quantitativa. Os trabalhos foram recuperados nas bases: Brapci, Peri e no Repositório da Federação Brasileira das Associações de Bibliotecários (FEBAB). O método utilizado foi a Revisão Sistemática de Literatura (RSL) e estatística descritiva resultaram na análise de trinta publicações. Concluiu-se que apesar do aumento no volume das publicações nos três últimos anos, considerado o período delimitado, a produção científica sobre aplicações da Inteligência Artificial na Ciência da Informação ainda é baixa e mostra as tendências de mais estudos em ‘organização e representação da informação’ e ‘machine larning’.

Abstract The advancements in Brain-Computer Interface (BCI) have substantially evolved people’s lives by enabling direct communication between the human brain and external peripheral devices. In recent years, the integration of machine larning (ML) and deep learning (DL) models have considerably imrpoved the performances of BCIs for decoding the motor imagery (MI) tasks. However, there still exist several limitations, e.g., extensive training time and high sensitivity to noises or outliers with those existing models, which largely hinder the rapid developments of BCIs. To address such issues, this paper proposes a novel extreme learning machine (ELM) based self-attention (E-SAT) mechanism to enhance subject-specific classification performances. Specifically, for E-SAT, ELM is employed both to imrpove self-attention module generalization ability for feature extraction and to optimize the model’s parameter initialization process. Meanwhile, the extracted features are also classified using ELM, and the end-to-end ELM based setup is used to evaluate E-SAT performances on different MI EEG signals. Extensive experiments with different datasets, such as BCI Competition III Dataset IV-a, IV-b and BCI Competition IV Datasets 1,2a,2b,3, are conducted to verify the effectiveness of proposed E-SAT strategy. Results show that E-SAT outperforms several state-of-the-art (SOTA) existing methods in subject-specific classification on all the datasets, with an average classification accuracy of 99.8%,99.1%,98.9%,75.8%, 90.8%, and 95.4%, being achieved for each datasets, respectively. The experimental results not only show outstanding performance of E-SAT in feature extractions, but also demonstrate that it helps achieves the best results among nine other robust ones. In addition, results in this study also demonstrate that E-SAT achieves exceptional performance in both binary and multi-class classification tasks, as well as for noisy and non-noisy datatsets.

Scientists have been working over years to assemble and accumulate data from biological sources to find solutions for many principal questions. Since a tremendous amount of data has been collected over the past and still increasing at an exponential rate, hence it now becomes unachievable for a human being alone to handle or analyze this data. Most of the data collection and maintenance is now done in digitalized format and hence requires an organization to have better data management and analysis to convert the vast data resource into insights to achieve their objectives. The continuous explosion of information both from biomedical and healthcare sources calls for urgent solutions. Healthcare data needs to be closely combined with biomedical research data to make it more effective in providing personalized medicine and better treatment procedures. Therefore, big data analytics would help in integrating large data sets for proper management, decision-making, and cost- effectiveness in any medical/healthcare organization. The scope of the thesis is to highlight the need for big data analytics in healthcare, explain data processing pipeline, and machine learning used to analyze big data.

La complexité croissante des écosystèmes agricoles et le La complexité croissante des écosystèmes agricoles et le besoin urgent de surveillance efficace de la santé des plantes rendent nécessaires des solutions technologiques avancées pour traiter les données textuelles. Située dans le cadre du projet BEYOND, cette thèse répond à ces besoins en améliorant les systèmes de reconnaissance d'entités nommées (REN) adaptés au domaine de la santé des plantes. Reconnaissant les limites des approches traditionnelles, cette recherche intègre des stratégies d'adaptation au domaine.La principale contribution de cette thèse réside dans le développement et l'affinement de méthodes destinées à améliorer l'adaptabilité des systèmes REN dans la reconnaissance d'informations liées à la santé des plantes, telles que les maladies végétales, les organismes nuisibles, les plantes et les lieux. En exploitant des techniques avancées d'apprentissage automatique, la thèse montre comment les systèmes REN peuvent être appliqués à la surveillance de la santé des plantes sans nécessiter d'adaptation explicite.Sur le plan méthodologique, la thèse adopte une approche double. D'une part, elle ajuste les modèles de langue grâce au masquage de mots-clés, focalisant le processus d'apprentissage sur le vocabulaire spécifique au domaine pour capturer les particularités linguistiques de la santé des plantes. D'autre part, elle améliore la reconnaissance des entités nommées grâce à l'intégration de représentations sémantiques obtenues à partir de descriptions textuelles des types d'entités. Cette méthode permet à l'algorithme de reconnaître des types d'entités non rencontrés durant l'apprentissage et de s'adapter facilement à de nouvelles applications. Cette méthodologie est ensuite appliquée aux données sur la santé des plantes, combinant les deux approches.Cette recherche contribue à l'avancement théorique dans le domaine de la REN et présente des implications pratiques, fournissant des outils susceptibles de conduire à une prise de décision plus informée face aux menaces phytosanitaires. Les orientations futures de ce travail incluent l'affinement des approches basées sur les lexiques, l'intégration de données multimodales et l'amélioration des définitions d'entités pour perfectionner davantage la précision et l'applicabilité des systèmes REN dans des domaines spécialisés tels que la santé des plantes
The increasing complexity of agricultural ecosystems and the urgent need for effective plant health monitoring necessitate advanced technological solutions for processing textual data. Situated within the BEYOND project, this thesis addresses these needs by advancing Named Entity Recognition (NER) systems tailored to the plant health domain. Considering the limitations of traditional NER approaches, this research innovates by integrating domain-specific adaptation strategies.The core contribution of this thesis is the development and refinement of methods to enhance the adaptability of NER systems in recognizing information related to plant health, such as diseases, pests, plants, and locations. By leveraging advanced machine learning techniques, the thesis demonstrates how NER systems can be applied to plant health monitoring without explicit adaptation.Methodologically, the thesis employs a dual approach. Firstly, it refines language models through Keyword Masking, focusing the training process on domain-relevant vocabulary to capture the specific linguistic features of the plant health domain. Secondly, it enhances entity recognition via semantic entity representations derived from textual descriptions of entity types. This approach enables the algorithm to identify entity types not seen during training, facilitating seamless adaptation to new applications. Finally, this methodology is applied to Plant Health data, combining both approaches for robust analysis.This research contributes theoretical advancements to the field of NER and offers practical implications for agricultural practices. It provides tools that can lead to more informed decision-making responses to plant health threats. Future directions for this work include refining lexicon-based approaches, integrating multimodal data, and enhancing the entity types definitions to further improve the precision and applicability of NER systems in specialized domains such as plant health