Depatment of Computer Science

Permanent URI for this community

https://kwasuspace.kwasu.edu.ng/handle/123456789/47

Browse

Now showing 1 - 8 of 8

A comparison of Boosting techniques for Classification of Microarray data
(Published by Faculty of Computing and Information Systems, University of Ilorin, 2023-04-13) Babatunde Ronke Seyi
Context: The advancements in technology, particularly microarrays, have played a pivotal role in enhancing crucial aspects within the domains of genomics and bioinformatics. These advancements have significantly contributed to the enhancement of illness diagnosis, evaluation of therapy response in patients, and advancements in cancer research. Microarray data often exhibits a substantial likelihood of encompassing extraneous and duplicative factors, hence introducing noise into the dataset. Consequently, the process of scrutinizing the data to identify significant patterns for diagnosis can be quite daunting when employing conventional statistical approaches. Numerous studies are currently being conducted to enhance the analysis of microarray data, with the aim of enhancing performance and prediction accuracy at an accelerated pace. Most of these earlier methods are limited in their predictive capacity and are characterised by high computational time and algorithm complexity. Objective: This research addresses some of these issues by implementing the classification of microarray data using Boosting algorithms. Method: Benchmarked on a publicly available dataset, the microarray data was cleaned, normalised and salient features carrying essential information were obtained. Three state of the art boosting algorithms; AdaBoost, Gradient Boost, and XGBoost were used in classifying the microarray data and the performance result of each was compared. Results: The experimental findings indicate that XGBoost demonstrates superior performance compared to other boosting approaches, with a classification accuracy rate of 98.18% and training time of 11seconds. Conclusions: The novelty of the experiment compared to earlier work is evident in the training time reported which is an information not frequently explicit in other report of findings.
A Kohonen Self Organizing Map (KSOM) Technique for Classification of Electrocardiogram (ECG) Signals.
(Published by Faculty of Computing and Informatics. Ladoke Akintola University of Technology, Ogbomoso. Oyo State., 2024-03-15) Babatunde Ronke Seyi
Electrocardiogram (ECG) signals are crucial in diagnosing cardiovascular diseases. Handling noisy ECG data, which is common in real-world situations makes accurate classification a critical task. Because ECG signals are faint and are quickly disrupted, classification accuracy can be poor, hence the need for improvement in the automatic ECG categorization system's recognition accuracy. The Kohonen Self Organising Map (KSOM) is known for its ability to cluster high-dimensional data in a low-dimensional space, hence its adoption in this research. The procedure employed include collection and pre-processing of diverse ECG data, including normal and abnormal cardiac rhythms. Inherent noise was removed from the data to ensure better-quality input data into the classification algorithm. A Kohonen SOM neural network (MiniSom model) was trained using the preprocessed ECG data. The KSOM organizes ECG signals into clusters on a topological map, preserving similarities and dissimilarities between different cardiac rhythms. Subsequently, the trained SOM serves as a reference model for classifying unseen ECG signals, indicating the corresponding cardiac rhythm. Benchmarked on two different dataset, evaluation of the classification performance of the technique was carried out. Cross-validation was done to assess the model's robustness and generalizability. Comparative analysis was conducted to measure the effectiveness and efficiency of the SOM-based approach against other common ECG signal classification techniques based on accuracy, precision, recall and fi-score. The result obtained shows that the average accuracy of 94.2%, precision of 83%, recall of 100% and f1-score of 91% achieved by the MiniSOM model outperformed the other models.
A predictive system for Parkinson Disease using Generative Adversarial network (GAN).
(Published by Federal University of Wukari (FUW), 2023-12-16) Babatunde Ronke Seyi
Parkinson's disease (PD) symptoms often overlap with those of other neurological disorders, making an early diagnosis difficult or even impossible. To tackle this problem, this study suggests a unique approach that makes use of Generative Adversarial Networks (GANs). Using a variety of datasets, GANs create synthetic medical data that includes PD-related clinical and demographic characteristics. The algorithm undergoes a rigorous training process to improve prediction accuracy. The generated data is assessed by a discriminator, which makes accurate PD predictions possible. Thorough measurements and statistical analysis verify the system's efficacy. The work demonstrates the revolutionary potential of GANs, especially in addressing data limitations for early Parkinson's disease diagnosis. The early PD detection efficacy of the technology is demonstrated by the very accurate predictive model. This study offers a sophisticated and useful method for early identification of Parkinson's disease (PD) by combining modern machine learning. It also promises improved patient outcomes by prompt neurology and healthcare interventions.
Comparative Analysis of Lempel-Ziv-Welch and Embedded Zero Tree Compression Algorithms on Kidney Images.
(Published by Faculty of Computing and Information Science. University of Ilorin. Special Issue on Computing and Communication Technologies, 2020-07-24) Babatunde Ronke Seyi
Image compression refers to the reduction of digital image data to store or transmit images with less rate of distortion. It is the reduction of the quantity of graphical data required to represent a digital image. Medical images play an important role in providing detailed information about patient injuries, fractures and other critical issues related to different disease and discomfort. Transmission of a reduced size of image data will help reduce computational burden as well as storage cost. Several compression algorithms have been adopted for use by numerous research on different types images, of which embedded zero tree (EZT) and Lempel-Ziv-Wetch (LZW) compression algorithms are most commonly used amongst others. The choice of a suitable compression algorithm for compression task on specific images has been an open issue, which this research seeks to look into. In this research, the compression capability of embedded zero tree (EZT) and Lempel-Ziv-Wetch (LZW) compression algorithms on Kidney images was compared. The performance of these algorithms was evaluated using peak signal noise ratio (PSNR), and compression ratio (CR). Effective compression implies that the resultant stream of image will be smaller than the original image size, without loss of vital content. Empirical results obtained shows that the LZW, having PSNR and CR of 97.28 and 2.85 respectively outperforms EZT with PSNR of 89.17 and CR of 1.82 when the same sets of images were subjected under the two compression algorithms. The results from this research suggest the usability of LZW for compression task involving kidney images. Similarly, lossless compression is recommended because loss of vital information content of medical images could lead to wrong diagnosis.
Investigating the Effect of Some Selected Distance Measures on the Performance of Query-By-Image-Content Over Mammogram Images
(2023-12-08) Babatunde Ronke Seyi
Introduction: Generating signatures of images and comparing them with those stored in a database for the purpose of retrieval of similar content is termed Query by image content. This art is helpful for detection of various diseases such as breast cancer, brain tumor and spine disorder. The image data are obtained through Computerized Tomography scan, Magnetic Resonance Imaging and mammogram, and are misunderstood in routine screening of mammograms despite the recent scientific improvements. Aim: This study aims to determine the most suitable distance measure for retrieval of similar images by experimenting state-of-the-art distance measures on mammographic images. Materials and Methods: The procedure was benchmarked on mini mammographic image analysis society (mini-MIAS) and breast cancer digital repository datasets. The experimental process includes thresholding and extraction of Region of Interest using gray level co-occurrence matrix. The extracted features were tested on Euclidean distance, Minkowski distance, Hamming distance, Mahalanobis distance, Cosine Similarity and Manhattan distance measures. The performance of the system on the distance measure was evaluated on the datasets to determine the distance metric that best identify abnormality in the samples. Results: The empirical results reveal that Mahalanobis distance measure outperforms the others in terms of retrieval time (1.26s and 1.14s) and minimal error (0.004 and 0.002) respectively for both the mini-MIAS dataset and the BCDR dataset, based on the similarity of images retrieved compared to queried images. Conclusion: Researchers can take appropriate decision on the choice of distance measure based on these findings
Towards Detection of Congenital Birth Defect in New born Using Autoencoder Neural Network
(Published by Faculty of Engineering and Technology, Adeleke University. Ede., 2023-04-12) Babatunde Ronke Seyi
Congenital birth defects are developmental issues occurring during fetal development. Such defects vary in severity and can affect structure, function, and metabolism of a fetus. Common examples include heart defects, cleft lip, and Down syndrome. Risk factors include maternal age, infections, medications, and genetics. Detecting these defects is crucial, given their impact on infant mortality. Medical imaging techniques allow healthcare professionals to visualize these defects and assess their severity, aiding in treatment planning and management. There has been several attempts to detecting birth defects using artificial intelligence and machine learning, some of which have yielded positive outcomes but are limited in terms of identifying and effectively predicting the defect, complexity in model interpretation and timely detection. These calls for continual research in this domain. This work implements an Autoencoder model for birth abnormality detection. The process involves data acquisition, preprocessing, feature extraction, model training and evaluation. Model evaluation was based on qualitative analysis and visualization. The experimental result shows the effectiveness and subsequent practicability of the Autoencoder model.
Towards prediction of cerebrospinal meningitis disease occurrence using Logistics Regression – a Web based Application
(Published by Federal University of Dutsin-Ma, Katsina State., 2023-04-18) Babatunde Ronke Seyi
Cerebrospinal meningitis (CSM) is characterized by acute severe infection of the central nervous system causing inflammation of the meninges with associated morbidity and mortality. The information about its symptoms, time and season of spread, most affected region, its fatality rate, type and how easily it causes major disabilities in patients can be modelled and utilized in its treatment, and prevention. This research uses data mining techniques to predict the occurrence of CSM in terms of those liable to be infected by the disease using feature information about the region and the patient. It encompasses data collection, preprocessing, exploration, algorithm training, prediction, and web hosting. The intention is to help in managing the resources needed for both treatment and prevention. The outcome of the research indicated that the proposed technique is viable for the task, considering the number of correct predictions that was reported when the application was deployed and tested.
Use of White Shark Optimization for Improving the Performance of Convolution Neural Network in Classification of Infected Citrus
(University of Ibadan Journal of Science and Logics in ICT Research., 2023-03-17) Babatunde Ronke Seyi
Citrus plant diseases are major causes of reduction in the production of citrus fruits and their usage. Early detection of the onset of the diseases is very important to curb and reduce its spread. A number of researches have been done on the detection and classification of the diseases, most of which have identified poor labelling of symptoms which result into improper classification. Some researchers have also experimented on the effectiveness of convolution neural network and other deep learning techniques, most of which results into a faster convergence but suffers from low accuracy, computational overhead and overfitting as fundamental issues. To reduce the effects of overfitting, this research developed a White Shark Optimization-Convolution Neural Network (WSO-CNN) technique to address the aforementioned problem by introducing a regularization strategy via feature selection which selects more useful and distinguishing features for classification. As a result, the developed technique was able to detect and classify various types of citrus fruit diseases and label them accordingly with low false positive rate, high sensitivity, specificity, increased accuracy and reduced recognition time, based on all the experiments performed with the dataset used in the research. Hence WSOCNN performed better than CNN in classifying citrus plant disease having a reduced FPR of 3.57%, 8.34%, 3.89% and 9.00% for black spot, greasy spot, canker and healthy/non healthy dataset respectively.

Browse

Browsing Depatment of Computer Science by Author "Babatunde Ronke Seyi"

Results Per Page

Sort Options