Visar Shehu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Visar Shehu has a total of 14 co-authors. Here are the top co-authors by number of shared publications:
- Adrian Besimi
South East European University
2 shared publications - Aleksandar Petkovski
South East European University
2 shared publications Labehat Kryeziu
Ph.D. Candidate, Faculty of Contemporary Sciences and Technologies South East European University, Tetovo, North Macedonia
2 shared publications- Merita Bakiji
South East European University
2 shared publications Adrian Besimi
South East European University
1 shared publication- Agron Chaushi
South East European University
1 shared publication
Latest Publications
- Open Access
- 15 March 2026
- Published by IntSys Research in Journal of Future Artificial Intelligence and Technologies
- Vol. 3 (1) , 19-35
A Lightweight Distance-Aware Loss for Thin Crack Segmentation in Building Facades under Limited-Data Conditions
Thin crack segmentation in building facades is particularly challenging under limited data conditions due to the extremely narrow geometry and weak contrast of crack structures. Conventional overlap-based loss functions, such as Binary Cross-Entropy (BCE) and Dice loss, optimize pixel-wise agreement but do not explicitly account for spatial boundary relevance, often resulting in fragmented predictions and geometric misalignment. This study introduces a lightweight distance-aware weighting extension of Dice loss designed to improve boundary alignment without modifying the network architecture. The proposed approach integrates Euclidean distance information derived from ground-truth masks to assign higher importance to prediction errors near crack boundaries while reducing the influence of distant background regions. The method is evaluated on a real-world dataset of 108 facade images (87 training and 21 validation) using a standard U-Net architecture under identical training conditions. Experimental results demonstrate a consistent reduction in geometric boundary error. The 95th percentile Hausdorff Distance (HD95) decreases from 230.08 px with BCE and 217.55 px with Dice loss to 148.28 px with the proposed distance-aware formulation, corresponding to reductions of approximately 81.8 px and 69.3 px, respectively. In addition, the proposed loss improves overlap-based metrics, achieving IoU@0.1 = 0.2844 and Dice = 0.4087 on the validation set. These results indicate that incorporating spatial distance information into the optimization objective improves geometric alignment and structural continuity of thin crack predictions. The findings suggest that integrating lightweight distance-aware weighting into conventional loss formulations can improve segmentation quality for thin structures in constrained-data scenarios while maintaining computational simplicity.
- 15 March 2026
- Published by IntSys Research in Journal of Future Artificial Intelligence and Technologies
- Vol. 3 (1) , 19-35
Open AccessA Lightweight Distance-Aware Loss for Thin Crack Segmentation in Building Facades under Limited-Data Conditions
Thin crack segmentation in building facades is particularly challenging under limited data conditions due to the extremely narrow geometry and weak contrast of crack structures. Conventional overlap-based loss functions, such as Binary Cross-Entropy (BCE) and Dice loss, optimize pixel-wise agreement but do not explicitly account for spatial boundary relevance, often resulting in fragmented predictions and geometric misalignment. This study introduces a lightweight distance-aware weighting extension of Dice loss designed to improve boundary alignment without modifying the network architecture. The proposed approach integrates Euclidean distance information derived from ground-truth masks to assign higher importance to prediction errors near crack boundaries while reducing the influence of distant background regions. The method is evaluated on a real-world dataset of 108 facade images (87 training and 21 validation) using a standard U-Net architecture under identical training conditions. Experimental results demonstrate a consistent reduction in geometric boundary error. The 95th percentile Hausdorff Distance (HD95) decreases from 230.08 px with BCE and 217.55 px with Dice loss to 148.28 px with the proposed distance-aware formulation, corresponding to reductions of approximately 81.8 px and 69.3 px, respectively. In addition, the proposed loss improves overlap-based metrics, achieving IoU@0.1 = 0.2844 and Dice = 0.4087 on the validation set. These results indicate that incorporating spatial distance information into the optimization objective improves geometric alignment and structural continuity of thin crack predictions. The findings suggest that integrating lightweight distance-aware weighting into conventional loss formulations can improve segmentation quality for thin structures in constrained-data scenarios while maintaining computational simplicity.
- Open Access
- 9 February 2026
- Published by International Association of Online Engineering (IAOE) in International Journal of Online and Biomedical Engineering (iJOE)
- Vol. 22 (02) , 142–153
Automatic Speech Recognition for Crisis Communication in the Albanian Language: Evaluating Whisper Turbo
This study evaluates the performance of the Whisper Turbo automatic speech recognition (ASR) model for crisis communication in the Albanian language. Applying a robust system such as Whisper Turbo will be very challenging because the stress and urgency of speaking in emergency situations will bring a rapid tempo and emotional intonation. We assess its accuracy and speed on two distinct datasets: a controlled corpus of formal, literary Albanian and a challenging corpus of emergency-style speech from first responders. The research uses word error rate (WER) and character error rate (CER) to quantify performance. We found a significant performance difference between the two conditions. The model achieved an average WER of 52.9% on formal Albanian but degraded to 63.5% on the dialectal and stressed speech. These results indicate a clear bias toward standardized language and highlight the model’s difficulty with non-standard pronunciations, emotional intonation, and specialized vocabulary. The findings underscore that while current multilingual ASR models can process low-resource languages such as Albanian, they are not yet suitable for deployment in critical emergency contexts without domain-specific fine-tuning. This work contributes an essential evaluation to the under-researched field of Albanian ASR and provides a foundation for developing more robust and reliable systems for crisis communication.
- 9 February 2026
- Published by International Association of Online Engineering (IAOE) in International Journal of Online and Biomedical Engineering (iJOE)
- Vol. 22 (02) , 142–153
Open AccessAutomatic Speech Recognition for Crisis Communication in the Albanian Language: Evaluating Whisper Turbo
This study evaluates the performance of the Whisper Turbo automatic speech recognition (ASR) model for crisis communication in the Albanian language. Applying a robust system such as Whisper Turbo will be very challenging because the stress and urgency of speaking in emergency situations will bring a rapid tempo and emotional intonation. We assess its accuracy and speed on two distinct datasets: a controlled corpus of formal, literary Albanian and a challenging corpus of emergency-style speech from first responders. The research uses word error rate (WER) and character error rate (CER) to quantify performance. We found a significant performance difference between the two conditions. The model achieved an average WER of 52.9% on formal Albanian but degraded to 63.5% on the dialectal and stressed speech. These results indicate a clear bias toward standardized language and highlight the model’s difficulty with non-standard pronunciations, emotional intonation, and specialized vocabulary. The findings underscore that while current multilingual ASR models can process low-resource languages such as Albanian, they are not yet suitable for deployment in critical emergency contexts without domain-specific fine-tuning. This work contributes an essential evaluation to the under-researched field of Albanian ASR and provides a foundation for developing more robust and reliable systems for crisis communication.
- 2 January 2026
- Published by Springer Nature
- p. 443-452
The Impact of Artificial Intelligence on the Workplace-Insight from North Macedonia’s IT Sector
- 2 January 2026
- Published by Springer Nature
- p. 443-452
The Impact of Artificial Intelligence on the Workplace-Insight from North Macedonia’s IT Sector
- 3 August 2025
- Published by Springer Nature
- p. 3-14
Improving Situational Awareness During Natural Disasters Using XR Technologies
- 3 August 2025
- Published by Springer Nature
- p. 3-14
Improving Situational Awareness During Natural Disasters Using XR Technologies
- Open Access
- 1 June 2025
- Published by Walter de Gruyter GmbH in SEEU Review
- Vol. 20 (1) , 92-109
Economic and Business Viability of It Outsourcing in North Macedonia
As in the global economy, in North Macedonia, IT Outsourcing is becoming an important economic element through which the country has become a very attractive and competitive destination at the same time. This study seeks to determine the contribution of the IT outsourcing sector to North Macedonia’s GDP and to identify the companies within North Macedonia that provide IT outsourcing services to the domestic market. This study analyzes the economic and business sustainability of IT outsourcing in North Macedonia, focusing on factors influencing sector development and industry challenges. This study uses the technique of collecting data on salaries, number of employees and other characteristics of companies from various economic and business platforms in the country. Incorporating this information into the study allows for strategic recommendations to be made to all stakeholders, including investors, policymakers, and IT professionals. These recommendations aim to achieve sustainable economic and business practices and to enhance the approach and strategy of outsourcing within business models.
- 1 June 2025
- Published by Walter de Gruyter GmbH in SEEU Review
- Vol. 20 (1) , 92-109
Open AccessEconomic and Business Viability of It Outsourcing in North Macedonia
As in the global economy, in North Macedonia, IT Outsourcing is becoming an important economic element through which the country has become a very attractive and competitive destination at the same time. This study seeks to determine the contribution of the IT outsourcing sector to North Macedonia’s GDP and to identify the companies within North Macedonia that provide IT outsourcing services to the domestic market. This study analyzes the economic and business sustainability of IT outsourcing in North Macedonia, focusing on factors influencing sector development and industry challenges. This study uses the technique of collecting data on salaries, number of employees and other characteristics of companies from various economic and business platforms in the country. Incorporating this information into the study allows for strategic recommendations to be made to all stakeholders, including investors, policymakers, and IT professionals. These recommendations aim to achieve sustainable economic and business practices and to enhance the approach and strategy of outsourcing within business models.
- 1 March 2025
- Published by Springer Nature
- p. 273-283
Anomaly Detection on Multivariate Sensing Time Series Data for Smart Aquaculture
- 1 March 2025
- Published by Springer Nature
- p. 273-283
Anomaly Detection on Multivariate Sensing Time Series Data for Smart Aquaculture
- Open Access
- 23 November 2023
- Published by Aleksander Moisiu University, Durres Albania in Interdisciplinary Journal of Research and Development
- Vol. 10 (3) , 7
Bert Based Named Entity Recognition for the Albanian Language
In this paper, we explore the application of the Bidirectional Encoder Representations from Transformers (BERT) model for Named Entity Recognition (NER) in the Albanian language. Despite the success of BERT in various NLP tasks across numerous languages, Albanian remains under-represented. Our approach leverages a transfer learning strategy with multilingual BERT, fine-tuned on a newly created, comprehensive Albanian language corpus. The corpus includes texts from various domains, thereby facilitating the identification of a broad range of named entities. We report on the challenges faced during corpus creation, model fine-tuning, and testing phases, such as dealing with dialectal variations and lack of existing resources for the Albanian language. This research not only contributes to the advancement of NER systems for low-resource languages, but also provides a robust foundation for further advancements in Albanian language processing. Received: 21 September 2023 / Accepted: 29 October 2023 / Published: 23 November 2023
- 23 November 2023
- Published by Aleksander Moisiu University, Durres Albania in Interdisciplinary Journal of Research and Development
- Vol. 10 (3) , 7
Open AccessBert Based Named Entity Recognition for the Albanian Language
In this paper, we explore the application of the Bidirectional Encoder Representations from Transformers (BERT) model for Named Entity Recognition (NER) in the Albanian language. Despite the success of BERT in various NLP tasks across numerous languages, Albanian remains under-represented. Our approach leverages a transfer learning strategy with multilingual BERT, fine-tuned on a newly created, comprehensive Albanian language corpus. The corpus includes texts from various domains, thereby facilitating the identification of a broad range of named entities. We report on the challenges faced during corpus creation, model fine-tuning, and testing phases, such as dealing with dialectal variations and lack of existing resources for the Albanian language. This research not only contributes to the advancement of NER systems for low-resource languages, but also provides a robust foundation for further advancements in Albanian language processing. Received: 21 September 2023 / Accepted: 29 October 2023 / Published: 23 November 2023
- Open Access
- 1 June 2023
- Published by Walter de Gruyter GmbH in SEEU Review
- Vol. 18 (1) , 1-16
Anomaly Detection on Univariate Sensing Time Series Data for Smart Aquaculture Using Deep Learning
Aquaculture plays a significant role in both economic development and food production. Maintaining an ecological environment with good water quality is essential to ensure the production efficiency and quality of aquaculture. Effective management of water quality can prevent abnormal conditions and contribute significantly to food security. Detecting anomalies in the aquaculture environment is crucial to ensure that the environment is maintained correctly to meet healthy and proper requirements for fish farming. This article focuses on the use of deep learning techniques to detect anomalies in water quality data in the aquaculture environment. Four deep learning anomaly detection techniques, including Autoencoder, Variational Autoencoder, Long-Short Term Memory Autoencoder, and Spectral-Residual Convolutional Neural Network, were analysed using multiple real-world sensor datasets collected from IoT aquaculture systems. Extensive experiments were conducted for temperature, dissolved oxygen, and pH parameters, and the evaluation analysis revealed that the Long-Short Term Memory Autoencoder anomaly detection method showed promising results in detecting anomalies for the temperature and oxygen datasets, while the Spectral-Residual Convolutional Neural Network demonstrated the best performance on the pH datasets.
- 1 June 2023
- Published by Walter de Gruyter GmbH in SEEU Review
- Vol. 18 (1) , 1-16
Open AccessAnomaly Detection on Univariate Sensing Time Series Data for Smart Aquaculture Using Deep Learning
Aquaculture plays a significant role in both economic development and food production. Maintaining an ecological environment with good water quality is essential to ensure the production efficiency and quality of aquaculture. Effective management of water quality can prevent abnormal conditions and contribute significantly to food security. Detecting anomalies in the aquaculture environment is crucial to ensure that the environment is maintained correctly to meet healthy and proper requirements for fish farming. This article focuses on the use of deep learning techniques to detect anomalies in water quality data in the aquaculture environment. Four deep learning anomaly detection techniques, including Autoencoder, Variational Autoencoder, Long-Short Term Memory Autoencoder, and Spectral-Residual Convolutional Neural Network, were analysed using multiple real-world sensor datasets collected from IoT aquaculture systems. Extensive experiments were conducted for temperature, dissolved oxygen, and pH parameters, and the evaluation analysis revealed that the Long-Short Term Memory Autoencoder anomaly detection method showed promising results in detecting anomalies for the temperature and oxygen datasets, while the Spectral-Residual Convolutional Neural Network demonstrated the best performance on the pH datasets.
- Open Access
- 1 June 2023
- Published by Walter de Gruyter GmbH in SEEU Review
- Vol. 18 (1) , 52-62
Pre-Training MLM Using Bert for the Albanian Language
Knowing that language is often used as a classifier of human intelligence and the development of systems that understand human language remains a challenge all the time (Kryeziu & Shehu, 2022). Natural Language Processing is a very active field of study, where transformers have a key role. Transformers function based on neural networks and they are increasingly showing promising results. One of the first major contributions to transfer learning in Natural Language Processing was the use of pre-trained word embeddings in 2010 (Joseph, Lev, & Yoshua, 2010). Pre-trained models like ELMo (Matthew, et al., 2018) and BERT (Delvin, et al., 2019) are trained on large corpora of unlabeled text and as a result learning from text representations has achieved good performance on many of the underlying tasks on datasets from different domains. Pre-training in the language model has proven that there has been an improvement in some aspects of natural language processing, based on the paper (Dai & Le, 2015). In present paper, we will pre-train BERT on the task of Masked Language Modeling (MLM) with the Albanian language dataset (alb_dataset) that we have created for this purpose (Kryeziu et al., 2022). We will compare two approaches: training of BERT using the available OSCAR dataset and using our alb_dataset that we have collected. The paper shows some discrepancies during training, especially while evaluating the performance of the model.
- 1 June 2023
- Published by Walter de Gruyter GmbH in SEEU Review
- Vol. 18 (1) , 52-62
Open AccessPre-Training MLM Using Bert for the Albanian Language
Knowing that language is often used as a classifier of human intelligence and the development of systems that understand human language remains a challenge all the time (Kryeziu & Shehu, 2022). Natural Language Processing is a very active field of study, where transformers have a key role. Transformers function based on neural networks and they are increasingly showing promising results. One of the first major contributions to transfer learning in Natural Language Processing was the use of pre-trained word embeddings in 2010 (Joseph, Lev, & Yoshua, 2010). Pre-trained models like ELMo (Matthew, et al., 2018) and BERT (Delvin, et al., 2019) are trained on large corpora of unlabeled text and as a result learning from text representations has achieved good performance on many of the underlying tasks on datasets from different domains. Pre-training in the language model has proven that there has been an improvement in some aspects of natural language processing, based on the paper (Dai & Le, 2015). In present paper, we will pre-train BERT on the task of Masked Language Modeling (MLM) with the Albanian language dataset (alb_dataset) that we have created for this purpose (Kryeziu et al., 2022). We will compare two approaches: training of BERT using the available OSCAR dataset and using our alb_dataset that we have collected. The paper shows some discrepancies during training, especially while evaluating the performance of the model.
- 1 January 2018
- Published by Springer Nature
- p. 194-202
Improving Employee Recruitment Through Data Mining
- 1 January 2018
- Published by Springer Nature
- p. 194-202
Improving Employee Recruitment Through Data Mining