Nebojsa Bacanin
37028223900
Publications - 6
Cloud spot instance price forecasting multi-headed models tuned using modified PSO
Publication Name: Journal of King Saud University Science
Publication Date: 2024-12-01
Volume: 36
Issue: 11
Page Range: Unknown
Description:
The increasing dependence and demands on cloud infrastructure have brought to light challenges associated with cloud instance pricing. The often unpredictable nature of demand as well as changing costs of supplying a reliable instance can leave companies struggling to appropriately budget to support a healthy cash flow while maintaining operating costs. This work explores the potential of multi-headed recurrent architectures to forecast cloud instance prices based on historical and instance data. Two architectures are explored, long short-term memory (LSTM) and gated recurrent unit (GRU) networks. A modified optimizer is introduced and tested on a publicly available Amazon elastic compute cloud dataset. The GRU model, enhanced by the proposed modified approach, had the most impressive outcomes with an MAE score of 0.000801. Results have undergone meticulous statistical validation with the best-performing models further analyzed using explainable artificial intelligence techniques to provide further insight into model reasoning and information on feature importance.
Open Access: Yes
Enhancing urban solar photovoltaic system performance evaluation through a disc spherical fuzzy aggregation framework
Publication Name: Journal of Computational Science
Publication Date: 2026-01-01
Volume: 93
Issue: Unknown
Page Range: Unknown
Description:
The integration of solar photovoltaic (PV) systems in urban environments promises great potential for sustainable energy applications. However, the unique characteristics of cities, the varieties of weather that occur at the place, and technology inefficiency make performance evaluation difficult. This paper sought to address the pressing need for a robust performance evaluation framework for urban solar PV systems by developing a disc spherical fuzzy aggregation framework. It develops basic algebraic aggregation operations in the framework of the disc spherical fuzzy set (D-SFSs), proving their completeness and describing their essential characteristics. These new operators conceived to operate on D-SFSs furnish theoretical robustness and provide the foundation for decisions made. A shining novel disc spherical fuzzy method is developed namely combinative distance-based assessment (CODAS) in D-SFS. A case study regarding the application of this model in the assessment of performance by urban solar PV systems is being conducted, thus proving the application aspect. Results come out positive in interpreting the decision-making dilemma and differences among several experts. This would, therefore, encourage various sectors to expand the use of D-SFSs in decision support systems and similar areas by showing how useful they can be in actual situations.
Open Access: Yes
Hospital Admission Classification of Cardiac Patients Utilizing Metaheuristics-Optimized Two Tier Framework
Publication Name: International Journal of Computational Intelligence Systems
Publication Date: 2026-12-01
Volume: 19
Issue: 1
Page Range: Unknown
Description:
Accurate evaluation of a cardiac patient’s risk at the point of hospital entry is critical for efficient triage and ensuring timely, suitable medical intervention. This study aims to forecast a range of clinical outcomes by leveraging admission data from a cardiac care unit, utilizing a refined and optimized machine learning approach. This research introduces a hybrid architecture that integrates convolutional neural networks (CNNs) with advanced machine learning classifiers, namely light gradient boosting machine (LightGBM) and categorical boosting (CatBoost), further enhanced through metaheuristic optimization techniques to maximize their performance. The proposed two-tiered design organizes feature extraction and final decision modeling into a coherent pipeline tailored for multi-class hospital admission classification. A comprehensive evaluation using a real-world hospital admission dataset demonstrates the framework’s effectiveness on a real-world, publicly available hospital admission dataset, supporting its utility for multi-class cardiac outcome prediction. Three experiments were conducted using publicly available datasets, where the best-performing models achieved a peak classification accuracy of 99.79%. Furthermore, explainable AI techniques were employed to interpret model predictions, offering actionable insights that can guide future data acquisition and strengthen the accurate classification of cardiac patients.
Open Access: Yes
A Modified Metaheuristic Optimization Approach for Forecasting the Lifecycle of Rechargeable Lithium-Ion Batteries
Publication Name: Smart Grids and Sustainable Energy
Publication Date: 2026-08-01
Volume: 11
Issue: 2
Page Range: Unknown
Description:
The global shift toward renewable energy is driven by the dual imperatives of rising energy demand and the need to reduce environmental harm caused by fossil fuels. However, renewables like wind and solar power pose unique challenges, particularly due to their intermittent generation and current limitations in energy storage technologies. Battery banks, commonly used to store surplus energy, degrade over time, making accurate forecasting of their remaining usable lifecycles critical for maintaining system reliability and efficiency. This study proposes a novel approach for forecasting battery health using an optimized long short-term memory (LSTM) network. To address the complexity of deep learning hyperparameter selection, a modified metaheuristic optimization algorithm is developed and integrated into a broader optimization framework aimed at improving model performance while minimizing overfitting. The method is benchmarked against several state-of-the-art optimizers, with results validated through comprehensive simulations and statistical analysis. This work contributes a scalable forecasting methodology, an effective optimization strategy, and interpretable results to support sustainable energy storage solutions.
Open Access: Yes
Optimizing robotic sensors in dynamic Industry 4.0 environments under a complex hesitant fuzzy soft model
Publication Name: Engineering Applications of Artificial Intelligence
Publication Date: 2026-02-15
Volume: 166
Issue: Unknown
Page Range: Unknown
Description:
The prevalence of autonomous robotic solutions has increased interest in multi-robot system connection and collaboration. Industry 4.0-driven manufacturing innovation will streamline production, improve quality, and sustain growth with robotic sensors. This paper presents a decision-making framework for optimizing robotic sensors in dynamic Industry 4.0 environments by introducing a Complex Hesitant Fuzzy Soft Set (CHFSS), as an extension of hesitant fuzzy sets and soft sets. The research aims to improve uncertainty and ambiguity management by increasing flexibility and authenticity with the complex plane. By assigning hesitating membership values in a multi-parameter complex unit plane, CHFSS is better equipped to demonstrate and handle challenging scenarios that occur throughout the decision-making process. A score function to sort the comparability issues between CHFSSs is introduced. Schweizer–Sklar’s operations and prioritized arithmetic and geometric aggregation operators are proposed. Their theorems and properties are developed and verified. The classical TODIM (acronym of the Portuguese expression TOmada de Decisão Interativa e Multicritério) approach is not able to deal with prioritized weight vectors. Thus, a novel CHFS-TODIM approach based on prioritized weight vectors is presented. The geometrical interpretation of the CHFS-TODIM approach under Schweizer–Sklar aggregation operators and prioritized weight vectors is depicted. A case study that focuses on the incorporation of robotic sensors is presented. Another goal of this case study is to show the approach’s precision, resilience, and flexibility. Additionally, it tests the viability and efficacy of the generated operators. The advantages and merits of the CHFS-TODIM approach are elaborated on.
Open Access: Yes
TB-CARE: a novel convolutional autoencoder-based tuberculosis classification system with enhanced EfficientNet
Publication Name: Peerj Computer Science
Publication Date: 2026-01-01
Volume: 12
Issue: Unknown
Page Range: Unknown
Description:
Tuberculosis (TB), caused by Mycobacterium tuberculosis (Mtb), is a significant health concern, with early detection playing a crucial role in reducing mortality. The similarity between TB and lung cancer poses a challenge for radiologists in avoiding misdiagnosis during chest X-ray inspections. In recent years, applying deep learning techniques to medical image classification has shown promise due to the various nature of datasets. In developing nations, screening examinations and early diagnosis would benefit substantially from an affordable, effective automated procedure. Thus, we present a novel deep learning architecture to facilitate the detection of TB by integrating multisource data from deep learning-based methods. The novel framework for TB classification using Convolutional Autoencoder with EfficientNet (TB-CARE) involves the utilisation of a convolutional autoencoder for feature extraction, an affinity propagation (AP) clustering method for selecting templates, and an enhanced EfficientNet (EEffNet) for classification. Extensive tests are performed on five freely accessible datasets. The results of our methodology surpassed those of previous approaches, demonstrating its practicality for real-world applications. By leveraging deep learning models within the ensemble method, TB classification achieves a notable accuracy in testing as 99% for the Belurus dataset and 83% on Shenzhen, 99% on the TBX dataset and 60% to 70% for the Montgomery dataset and National Institutes of Health (NIH), where various classes with high imbalance affect the model's performance.
Open Access: Yes
