Találati lista | Tudóstér

001-es BibID:	BIBFORM124274
Első szerző:	Alshuwaili, Dhafer Gheni Honi (Informatics)(PhD)
Cím:	A comparative study of pre-trained models in breast ultrasound image segmentation / Dhafer G. Honi, Mohammed Nsaif, Laszlo Szathmary, Szilvia Szeghalmy
Dátum:	2024
Megjegyzések:	In recent years, several deep learning architectures have emerged achieving impressive results in breast ultrasound image segmentation, despite the fact that the problem itself is challenging, because of the variation in lesion size and unequal distribution of intensity in the lesion area. Many of these methods were trained and evaluated on a specific dataset, the Breast Ultrasound Images (BUSI), as it was one of the first publicly available datasets in the field with expert annotations. However, recently, problems with the dataset have been discovered. We conducted our research to estimate, through a few selected methods, the extent to which problems with the dataset make the performance values reported in recent years unreliable. To achieve this, the selected procedures were trained and evaluated along the same methodology on the original and the cleaned datasets. Our results indicate that results related to the BUSI collection should be treated with serious caution.
ISBN:	979-8-3503-8788-9
Tárgyszavak:	Műszaki tudományok Informatikai tudományok előadáskivonat könyvrészlet segmentation deep learning breast cancer ultrasound images
Megjelenés:	2024 IEEE 3rd Conference on Information Technology and Data Science (CITDS) Proceedings /András Hajdu. - p. 81-86. -
További szerzők:	Nsaif, Mohammed (informatics) Szathmáry László (1977-) (programtervező-informatikus) Szeghalmy Szilvia (1984-) (programtervező matematikus)
Internet cím:	Intézményi repozitóriumban (DEA) tárolt változat
Borító:
Saját polcon:

001-es BibID:	BIBFORM132970
035-os BibID:	(WOS)001612830300001 (Scopus)105020809538
Első szerző:	Dawod, Abdaljbbar Babiker Abdaljbbar (Ph.D. student)
Cím:	Value-at-Risk Estimation in the Tadawul Exchange : A Comparison of Recurrent Neural Networks and GARCH Models / Abdaljbbar B. A. Dawod, Mohammed Nsaif, Dhafer G. Honi, Ismail H. Abdi, Husam A. Neamah
Dátum:	2025
ISSN:	2772-9419
Megjegyzések:	Value at Risk (VaR) and Expected Shortfall (ES) are two of the most established measures of market risk, yet both can lose reliability in volatile and nonlinear markets. This study examines how econometric models compare with modern recurrent neural networks in forecasting risk for the Saudi stock market. Using 4,340 daily closing prices of the Tadawul All Share Index (TASI) from March 2007 to July 2024, the analysis applies a sliding-window framework. Forecast accuracy is assessed through Root Mean Square Error (RMSE) and information criteria, while VaR and ES forecasts are validated using unconditional and conditional coverage tests. The study makes three key contributions: (i) it systematically contrasts GARCH specifications with recurrent models (LSTM, BiLSTM, GRU, BiGRU) in a long-horizon dataset, (ii) it extends backtesting of VaR and ES across 17 years of data, and (iii) it provides practical insights for combining econometric and machine-learning methods in emerging markets. Results show that BiGRU achieved the highest predictive accuracy, while GRU offered the best balance between accuracy and parsimony. Heavy-tailed GARCH models produced the most reliable tail-risk forecasts, consistently passing tests at the 95% and 99% confidence levels. Although recurrent networks were less accurate at extreme quantiles, they effectively captured nonlinear patterns and volatility clustering, making them a valuable complement for practical risk forecasting.
Tárgyszavak:	Műszaki tudományok Informatikai tudományok idegen nyelvű folyóiratközlemény külföldi lapban folyóiratcikk Tadawul GARCH LSTM BiLSTM GRU BiGRU VaR ES
Megjelenés:	Systems and Soft Computing. - 7 (2025), p. 1-14. -
További szerzők:	Nsaif, Mohammed (informatics) Alshuwaili, Dhafer Gheni Honi (1991-) (Informatics)(PhD) Abdi, Ismail H. Neamah, Husam A. (1990-) (mérnök)
Internet cím:	Intézményi repozitóriumban (DEA) tárolt változat DOI Szerző által megadott URL
Borító:
Saját polcon: