Faculty of Electrical Engineering and Information Technologies

Permanent URI for this communityhttps://repository.ukim.mk/handle/20.500.12188/10

Browse

Settings

Has files: YesSubject: search.filters.subject.Machine Learning

Search Results

Now showing 1 - 6 of 6
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    Classifying Power Quality Disturbances in Noisy Conditions using Machine Learning
    (The Jozhef Stefan Institute, 2019-10)
    Velichkovska, Bojana
    ;
    Markovska, Marija
    ;
    Gjoreski, Hristijan
    ;
    When ensuring high-quality power supply of the power grid it is of the upmost importance to correctly detect and classify any power quality (PQ) disturbance. Selecting the most relevant features is very important in the process of training a genera machine learning model. Therefore, we analyze the power signals and extract information from them, and then select the most significant features. Additionally, an effective classification model is required. In this study we apply grid search throughout the features sets on one side, and the classification algorithms on the side. This way, we determine the most effective combination of an algorithm and feature set for classification of power quality disturbances.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    Investigating Presence of Ethnoracial Bias in Clinical Data using Machine Learning
    (2021-09)
    Velichkovska, Bojana
    ;
    Gjoreski, Hristijan
    ;
    ;
    ;
    Celi, Leo Anthony
    An important target for machine learning research is obtaining unbiased results, which require addressing bias that might be present in the data as well as the methodology. This is of utmost importance in medical applications of machine learning, where trained models should be unbiased so as to result in systems that are widely applicable, reliable and fair. Since bias can sometimes be introduced through the data itself, in this paper we investigate the presence of ethnoracial bias in patients’ clinical data. We focus primarily on vital signs and demographic information and classify patient ethnoraces in subsets of two from the three ethnoracial groups (African Americans, Caucasians, and Hispanics). Our results show that ethnorace can be identified in two out of three patients, setting the initial base for further investigation of the complex issue of ehtnoracial bias.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    Mobile Edge Computing services with QoS support for beyond 5G Networks – Use Cases
    (2021-09)
    Nunev, David
    ;
    Shuminoski, Tomislav
    ;
    Velichkovska, Bojana
    ;
    This paper presents a novel research in intelligent multi-access QoS mobile edge computing (MEC) for beyond 5G services. Also, the improved advanced QoS model and architecture for beyond 5G systems and services are proposed. The proposed model combines the most powerful features of both Cloud and Edge computing, independent from any existing and future Radio Access Technology, leading to high performance utility networks with high QoS provisioning for any used multimedia modern service over present and future mobile and wireless networks and systems. Moreover, the proposed architecture will allow applications and network services to be executed at the edge part of the network, giving lower end-to-end delay for the end-user services and applications. Finally, this paper gives an overview of the existing Mobile Edge Computing technologies and several use cases. Undoubtedly, MEC is an innovative network paradigm going beyond 5G to cater for the unprecedented growth of computation demands and the ever increasing computation quality of user experience requirements.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    Image Segmentation as an Instrument for Setting Attention Regions in Convolutional Neural Networks for Bias Detection Purposes
    (2023)
    Velichkovska, Bojana
    ;
    Efnusheva, Danijela
    ;
    ;
    Convolutional neural networks (CNNs) are constantly being used for medical image processing with increased application in publicly available datasets and are later being actively applied in medical practice. Therefore, since patient lives are at stake, it is important that the functionality of the neural network is beyond reproach. In this paper, due to dataset availability, we present two lung segmentation approaches using traditional image processing and deep learning methodologies; these approaches can later be used to focus a CNN for image segmentation and classification tasks, with implementations spanning everything from disease diagnosis to demographic and bias analysis. The aim of this paper is to provide a framework for segmentation in medical images of the chest cavity, as a way of applying attention regions and localizing sources of bias in images. Both of the proposed segmentation tools, the traditional image approach using computer tomography scans and the CNN applied to chest X-rays, provide excellent lung segmentation comparable to popular methods in the image processing sphere. This allows for an all-encompassing application of the developed methodology regardless of different image formats, therefore making it widely applicable in setting attention regions for CNNs.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    Demographic Bias in Medical Datasets for Clinical AI
    (2023)
    Velichkovska, Bojana
    ;
    Petrushevska, Sandra
    ;
    Runcheva, Bisera
    ;
    Numerous studies have detailed instances of demographic bias in medical data and artificial intelligence (AI) systems used in medical setting. Moreover, these studies have also shown how these biases can significantly impact the access to and quality of care, as well as quality of life for patients belonging in certain under-represented groups. These groups are then being marginalised because of stigma based on demographic information such as race, gender, age, ability, and so on. Since the performance of AI models is highly dependent on the quality of data used to train the algorithms, it is a necessary precaution to analyse any potential bias inadvertently existent in the data, in order to mitigate the consequences of using biased data in creating medical AI systems. For that reason, we propose a machine learning (ML) analysis which receives patient biosignals as input information and analyses them for two types of demographic bias, namely gender and age bias. The analysis is performed using several ML algorithms (Logistic Regression, Decision Trees, Random Forest, and XGBoost). The trained models are evaluated with a holdout technique and by observing the confusion matrixes and the classification reports. The results show that the models are capable of detecting bias in data. This makes the proposed approach one way to identify bias in data, especially throughout the process of building AI-based medical systems. Consequently, the proposed pipeline can be used as a mitigation technique for bias analysis in data.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    Machine Learning-Based Forecasting of Bitcoin Price Movements
    (2024)
    Angelovski, Darko
    ;
    Velichkovska, Bojana
    ;
    ;
    Efnusheva, Daniela
    ;
    In the volatile realm of cryptocurrency markets, this research explores the intricate dance of Bitcoin price dynamics through the lens of machine learning. Employing a multifaceted approach, we harness the power of Long Short-Term Memory (LSTM) networks, Gradient Boosting, LightGBM (LGBM) Regressor, and Random Forest algorithms to unravel the complexities of price movements. We perform a comprehensive analysis, and observe patterns and dependencies within historical data at hour-long intervals in the last 30 and 45 days, by using a holdout technique with 80% of the data used for training and 20% used for testing. We evaluate the models using four standard regression metrics. The training data incorporates a diverse range of features capturing hourly trends, day-of-the-week variations, and the correlation between opening and closing prices. Our study delves into the ability for forecasting Bitcoin price movements using ensemble algorithms and LSTM. The results show best performance for the LSTM models, especially when trained on longer training intervals. Namely, our LSTM model obtains R2 of 0.98 when trained on 30 days and 0.99 when trained on 45 days. In comparison, the ensemble methods show volatility and lower predictive ability.