Kate Takyi
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Kate Takyi has a total of 43 co-authors. Here are the top co-authors by number of shared publications:
- Rose-Mary Owusuaa Mensah Gyening
Kwame Nkrumah University of Science and Technology
11 shared publications Amandeep Bagga
2 shared publications- Eshun Michael
Brock University
2 shared publications - James Ben Hayfron-Acquah
Faculty of Physical and Computational Sciences, Kwame Nkrumah University of Science and Technology (KNUST)
2 shared publications Kwabena Owusu-Agyemang
2 shared publicationsPeter Appiahene
University of Energy and Natural Resources, Department of Computer Science and Informatics, Ghana
2 shared publications
Latest Publications
- Open Access
- 2 November 2026
- Published by Oxford University Press (OUP) in Oxford Open Digital Health
MOSQMIXERNET: A CUSTOM DEEP LEARNING ARCHITECTURE FOR EARLY-STAGE MOSQUITO CLASSIFICATION AND GENUS-LEVEL IDENTIFICATION
Mosquito-borne diseases such as malaria, dengue, zika, and chikungunya impose a disproportionate heavy public health burden on Africa. While artificial intelligence (AI) tools have been applied in species identification on the adult mosquitoes, the early aquatic stages have been largely ignored. Yet, these early stages are critical since vector control strategies often target larvae and pupae to break the transmission cycle before mosquitoes mature into biting adults. Moreover, the identification of these specimens is prone to human error by experts due to the striking similarity in morphological features between larval and pupal stages across species. This limitation delays surveillance and increases the risk of misclassification, potentially compromising vector control efforts. In response, an Africa-based curated dataset is introduced, encompassing the larval and pupal stages of Aedes, Anopheles, and Culex species, along with non-mosquito specimens of similar morphology. This dataset helps address the underrepresentation of African samples in global repositories and provides a rare contribution to pupal-stage classification, a category largely overlooked in most classification tasks. MosqMixerNet is introduced as a lightweight deep learning model designed for image-based classification of mosquito early stages. The proposed MosqMixerNet balances compactness and accuracy with only 169 666 trainable parameters and was benchmarked against MobileNetV2, DenseNet121, and NASNetMobile. For genus-level identification, it achieved 99.37% accuracy, surpassing NASNetMobile (97.30%), MobileNetV2 (97.26%), and DenseNet121 (80.39%). The work advances computer vision for public health and tropical disease control, and provides a foundation for integration into digital vector monitoring platforms in endemic settings.
- 2 November 2026
- Published by Oxford University Press (OUP) in Oxford Open Digital Health
Open AccessMOSQMIXERNET: A CUSTOM DEEP LEARNING ARCHITECTURE FOR EARLY-STAGE MOSQUITO CLASSIFICATION AND GENUS-LEVEL IDENTIFICATION
Mosquito-borne diseases such as malaria, dengue, zika, and chikungunya impose a disproportionate heavy public health burden on Africa. While artificial intelligence (AI) tools have been applied in species identification on the adult mosquitoes, the early aquatic stages have been largely ignored. Yet, these early stages are critical since vector control strategies often target larvae and pupae to break the transmission cycle before mosquitoes mature into biting adults. Moreover, the identification of these specimens is prone to human error by experts due to the striking similarity in morphological features between larval and pupal stages across species. This limitation delays surveillance and increases the risk of misclassification, potentially compromising vector control efforts. In response, an Africa-based curated dataset is introduced, encompassing the larval and pupal stages of Aedes, Anopheles, and Culex species, along with non-mosquito specimens of similar morphology. This dataset helps address the underrepresentation of African samples in global repositories and provides a rare contribution to pupal-stage classification, a category largely overlooked in most classification tasks. MosqMixerNet is introduced as a lightweight deep learning model designed for image-based classification of mosquito early stages. The proposed MosqMixerNet balances compactness and accuracy with only 169 666 trainable parameters and was benchmarked against MobileNetV2, DenseNet121, and NASNetMobile. For genus-level identification, it achieved 99.37% accuracy, surpassing NASNetMobile (97.30%), MobileNetV2 (97.26%), and DenseNet121 (80.39%). The work advances computer vision for public health and tropical disease control, and provides a foundation for integration into digital vector monitoring platforms in endemic settings.
- 13 May 2026
- Published by Taylor & Francis in Disability and Rehabilitation: Assistive Technology
- Vol. 21 (7) , 3438-3458
A systematic review of machine learning techniques for real-time speech and noise classification in hearing aids
Introduction: Hearing loss significantly impairs speech comprehension in noisy environments, creating major communication challenges for individuals with hearing impairment. Modern hearing aids increasingly rely on intelligent systems capable of real-time speech and noise classification to enhance speech intelligibility while suppressing background noise. Researchers have explored signal processing, machine learning, and deep learning approaches to improve classification accuracy, adaptability, and performance in complex acoustic environments. Methods: A systematic review was conducted following Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020 guidelines to identify peer-reviewed studies on machine learning-based speech and noise classification for hearing aids published between 2004 and 2024. Searches were conducted across major academic databases using structured keywords. Studies were screened using predefined inclusion and exclusion criteria, and quality assessment was performed using a structured scoring framework. Due to methodological heterogeneity, findings were synthesised narratively. Results: The review found that machine learning and deep learning approaches consistently outperform traditional signal processing techniques in speech and noise classification tasks. Deep learning architectures, particularly convolutional neural networks, recurrent neural networks, long short-term memory networks, and hybrid models, demonstrated improved robustness, adaptability, and classification performance in noisy and dynamic acoustic environments. Emerging trends include lightweight architectures, model compression, ensemble learning, and context-aware systems designed to support real-time deployment in hearing aids. Conclusion: Machine learning and deep learning techniques have significantly advanced speech and noise classification in hearing aids. However, challenges related to latency, computational efficiency, energy consumption, and real-world deployment remain. Future research should prioritise lightweight, adaptive, and user-centred hearing aid technologies.
- 13 May 2026
- Published by Taylor & Francis in Disability and Rehabilitation: Assistive Technology
- Vol. 21 (7) , 3438-3458
A systematic review of machine learning techniques for real-time speech and noise classification in hearing aids
Introduction: Hearing loss significantly impairs speech comprehension in noisy environments, creating major communication challenges for individuals with hearing impairment. Modern hearing aids increasingly rely on intelligent systems capable of real-time speech and noise classification to enhance speech intelligibility while suppressing background noise. Researchers have explored signal processing, machine learning, and deep learning approaches to improve classification accuracy, adaptability, and performance in complex acoustic environments. Methods: A systematic review was conducted following Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020 guidelines to identify peer-reviewed studies on machine learning-based speech and noise classification for hearing aids published between 2004 and 2024. Searches were conducted across major academic databases using structured keywords. Studies were screened using predefined inclusion and exclusion criteria, and quality assessment was performed using a structured scoring framework. Due to methodological heterogeneity, findings were synthesised narratively. Results: The review found that machine learning and deep learning approaches consistently outperform traditional signal processing techniques in speech and noise classification tasks. Deep learning architectures, particularly convolutional neural networks, recurrent neural networks, long short-term memory networks, and hybrid models, demonstrated improved robustness, adaptability, and classification performance in noisy and dynamic acoustic environments. Emerging trends include lightweight architectures, model compression, ensemble learning, and context-aware systems designed to support real-time deployment in hearing aids. Conclusion: Machine learning and deep learning techniques have significantly advanced speech and noise classification in hearing aids. However, challenges related to latency, computational efficiency, energy consumption, and real-world deployment remain. Future research should prioritise lightweight, adaptive, and user-centred hearing aid technologies.
- 14 April 2026
- Published by Scifiniti Publishing in Computing&AI Connect
Attention-Based LSTM for Sign Language Recognition Leveraging Spatial-Temporal Keypoint
Sign language is a crucial means of communication for the Deaf and hard-of-hearing communities. Most individuals find it challenging to communicate with the deaf when they try to do so without an interpreter. The advancement in technology, computer vision, and deep learning approaches provides a different approach to tackling the problem. Literature indicates that the unique nature of Ghanaian Sign Language (GSL) has been understudied due to a lack of large and publicly available datasets, as well as limited research on the use of landmark keypoints for computational research on GSL. This study curated a large video-based dataset, AkwaabaSign, that reflects the indigenous nature of the GSL. The study employed two different baseline models to assess the dataset: an Attention-enhanced LSTM model and a ConvLSTM model, which extracted and specially normalized the keypoints using Mediapipe. With this approach, the attention-enhanced LSTM achieved a test accuracy of 94.69%, with balanced performance metrics of 93.32% precision, 92.70% recall, and 92.66% F1-score. The ConvLSTM achieved 90.28% accuracy, lagging behind the attention-enhanced LSTM. The study fulfils the aim of producing a large dataset for sign language recognition, provides a specialized normalization process for dataset processing, and establishes a base model for the practical use of the dataset. The proposed model also outperforms some other algorithms in the domain of sign language computational works in GSL. The study aims to expand the dataset to the sentence level and develop continuous GSL recognition.
- 14 April 2026
- Published by Scifiniti Publishing in Computing&AI Connect
Attention-Based LSTM for Sign Language Recognition Leveraging Spatial-Temporal Keypoint
Sign language is a crucial means of communication for the Deaf and hard-of-hearing communities. Most individuals find it challenging to communicate with the deaf when they try to do so without an interpreter. The advancement in technology, computer vision, and deep learning approaches provides a different approach to tackling the problem. Literature indicates that the unique nature of Ghanaian Sign Language (GSL) has been understudied due to a lack of large and publicly available datasets, as well as limited research on the use of landmark keypoints for computational research on GSL. This study curated a large video-based dataset, AkwaabaSign, that reflects the indigenous nature of the GSL. The study employed two different baseline models to assess the dataset: an Attention-enhanced LSTM model and a ConvLSTM model, which extracted and specially normalized the keypoints using Mediapipe. With this approach, the attention-enhanced LSTM achieved a test accuracy of 94.69%, with balanced performance metrics of 93.32% precision, 92.70% recall, and 92.66% F1-score. The ConvLSTM achieved 90.28% accuracy, lagging behind the attention-enhanced LSTM. The study fulfils the aim of producing a large dataset for sign language recognition, provides a specialized normalization process for dataset processing, and establishes a base model for the practical use of the dataset. The proposed model also outperforms some other algorithms in the domain of sign language computational works in GSL. The study aims to expand the dataset to the sentence level and develop continuous GSL recognition.
- Open Access
- 1 April 2026
- Published by Elsevier in SSRN Electronic Journal
A Framework for Secure Microservice Architecture for Cloud- based Applications
Microservices architecture is becoming more popular, which presents issues for maintainability, scalability and security. This paper suggests a unique framework that combines Docker, Kubernetes, and role-based access control (RBAC) to create safe microservice architectures. The framework makes use of Docker for easy deployment and management, Kubernetes for effective orchestration and scaling, and RBAC for granular access control. With Kubernetes automating activities like load balancing and scaling, Docker simplifying package development, deployment, and management are made simpler. The paradigm also helps with fault tolerance with segregated services, observability with integrated monitoring tools, and maintainability with modular microservices. Major achievements consist of enhanced security by limiting user rights based on responsibilities. The study proposes a secure ecosystem of microservices that comprises Docker, Kubernetes, and RBAC. The implementation also showed more reasonable deployment times for the individual microservices, ranging from 2 minutes 33 seconds to 3 minutes 54 seconds, depending on the resources available and the network space. The framework use was beneficial to RBAC, because the principles of the system were used to provide least privilege access and granularity of permissions. It also demonstrated better fault tolerance and resource management and therefore qualifies as an implementable solution to secure and elastic and modular cloud-based microservice applications. The advances of the proposed framework showed a decrease in unauthorized access, enforcement of the policy, and minimal increments in the latency overhead.
- 1 April 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessA Framework for Secure Microservice Architecture for Cloud- based Applications
Microservices architecture is becoming more popular, which presents issues for maintainability, scalability and security. This paper suggests a unique framework that combines Docker, Kubernetes, and role-based access control (RBAC) to create safe microservice architectures. The framework makes use of Docker for easy deployment and management, Kubernetes for effective orchestration and scaling, and RBAC for granular access control. With Kubernetes automating activities like load balancing and scaling, Docker simplifying package development, deployment, and management are made simpler. The paradigm also helps with fault tolerance with segregated services, observability with integrated monitoring tools, and maintainability with modular microservices. Major achievements consist of enhanced security by limiting user rights based on responsibilities. The study proposes a secure ecosystem of microservices that comprises Docker, Kubernetes, and RBAC. The implementation also showed more reasonable deployment times for the individual microservices, ranging from 2 minutes 33 seconds to 3 minutes 54 seconds, depending on the resources available and the network space. The framework use was beneficial to RBAC, because the principles of the system were used to provide least privilege access and granularity of permissions. It also demonstrated better fault tolerance and resource management and therefore qualifies as an implementable solution to secure and elastic and modular cloud-based microservice applications. The advances of the proposed framework showed a decrease in unauthorized access, enforcement of the policy, and minimal increments in the latency overhead.
- Open Access
- 10 February 2026
- Published by Elsevier in SSRN Electronic Journal
Zero Trust and Event-Driven Access Control for Ransomware Mitigation: A Review Paper
Cloud computing, the Internet of Things (IoT), and distributed service architectures have revolutionized the way enterprises control access and identity services, but have also enlarged the threat landscape to include ransomware and insider attacks. Traditional perimeter-driven security systems have been unable to deal with polymorphic, fileless, and adaptive malware. The paper presents a review of Zero Trust Architecture (ZTA) and Staged Event-Driven Access Control (SEDAC) as intersectional frameworks of adaptive cybersecurity. This paper examines the working mechanisms and architectural layers of the major frameworks, as well as the outcomes of their empirical work. Some analyses emphasize the efficiency of defense, the complexity of implementation, and unresolved problems. The review concludes that, in the context of dynamic and context-aware security postures in current cloud and IoT systems, combining Zero Trust continuous verification with SEDAC behavioral staging can be a relevant approach.
- 10 February 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessZero Trust and Event-Driven Access Control for Ransomware Mitigation: A Review Paper
Cloud computing, the Internet of Things (IoT), and distributed service architectures have revolutionized the way enterprises control access and identity services, but have also enlarged the threat landscape to include ransomware and insider attacks. Traditional perimeter-driven security systems have been unable to deal with polymorphic, fileless, and adaptive malware. The paper presents a review of Zero Trust Architecture (ZTA) and Staged Event-Driven Access Control (SEDAC) as intersectional frameworks of adaptive cybersecurity. This paper examines the working mechanisms and architectural layers of the major frameworks, as well as the outcomes of their empirical work. Some analyses emphasize the efficiency of defense, the complexity of implementation, and unresolved problems. The review concludes that, in the context of dynamic and context-aware security postures in current cloud and IoT systems, combining Zero Trust continuous verification with SEDAC behavioral staging can be a relevant approach.
- Open Access
- 1 January 2026
- Published by Institute of Advanced Engineering and Science in Indonesian Journal of Electrical Engineering and Computer Science
- Vol. 41 (1) , 320-330
Deep-fuzzy personalisation framework for robot-assisted learning for children with autism
Research exploring the efficacy of robots in autism therapy has predominantly relied on the Wizard-of-Oz method, where robots execute predetermined behaviours. However, this approach is constrained by its heavy reliance on human intervention. To address this limitation, we introduce a novel deep-fuzzy personalization framework for social robots to enhance adaptability in interactions with autistic children. This framework incorporates a deep learning model called singleshot emotion detector (SED) with a mean average precision of 93% and a fuzzy-based engagement prediction engine, utilizing factors such as scores, IQ levels, and task complexity to estimate the engagement of autistic children during robot interactions. Implemented on the humanoid robot RoCA, our study assesses the impact of this personalization approach on learning outcomes in interactions with Ghanaian autistic children. Statistical analysis, specifically Mann Whitney tests (U=3.0, P=0.012), demonstrates the significant improvement in learning gains associated with RoCA's adoption of the deep fuzzy approach.
- 1 January 2026
- Published by Institute of Advanced Engineering and Science in Indonesian Journal of Electrical Engineering and Computer Science
- Vol. 41 (1) , 320-330
Open AccessDeep-fuzzy personalisation framework for robot-assisted learning for children with autism
Research exploring the efficacy of robots in autism therapy has predominantly relied on the Wizard-of-Oz method, where robots execute predetermined behaviours. However, this approach is constrained by its heavy reliance on human intervention. To address this limitation, we introduce a novel deep-fuzzy personalization framework for social robots to enhance adaptability in interactions with autistic children. This framework incorporates a deep learning model called singleshot emotion detector (SED) with a mean average precision of 93% and a fuzzy-based engagement prediction engine, utilizing factors such as scores, IQ levels, and task complexity to estimate the engagement of autistic children during robot interactions. Implemented on the humanoid robot RoCA, our study assesses the impact of this personalization approach on learning outcomes in interactions with Ghanaian autistic children. Statistical analysis, specifically Mann Whitney tests (U=3.0, P=0.012), demonstrates the significant improvement in learning gains associated with RoCA's adoption of the deep fuzzy approach.
- Open Access
- 9 September 2025
- Published by Elsevier in Data in Brief
- Vol. 62, 112045
MeatScan: An image dataset for machine learning-based classification of fresh and spoiled cow meat
- 9 September 2025
- Published by Elsevier in Data in Brief
- Vol. 62, 112045
Open AccessMeatScan: An image dataset for machine learning-based classification of fresh and spoiled cow meat
- Open Access
- 31 August 2025
- Published by AJIT - E Academic Journal of Information Technology in AJIT-e: Academic Journal of Information Technology
- Vol. 16 (3) , 207-231
Advanced Mobile Money Fraud Detection Using CNN-BiLSTM and Optimized SGD with Momentum
The accelerated adoption of mobile money systems has significantly increased fraudulent activity, compromising their security and trustworthiness. This research presents an enhanced method for detecting mobile money fraud by modifying a CNN-BiLSTM model with momentum using Stochastic Gradient Descent (SGD). We computed salient features from transaction data using a pre-processed hybrid CNN-BiLSTM model and trained the model to identify trends in the data that included geographical and temporal aspects. The model performed remarkably using industry-standard testing approaches: an F1 score of 0.9928, precision of 0.9927, accuracy of 0.9928, and recall of 0.9929. The proposed model can identify dishonesty and has a low false positive rate. According to the study, the model improves feature selection and incorporates various optimization techniques, making it more flexible and suitable for different mobile money systems.
- 31 August 2025
- Published by AJIT - E Academic Journal of Information Technology in AJIT-e: Academic Journal of Information Technology
- Vol. 16 (3) , 207-231
Open AccessAdvanced Mobile Money Fraud Detection Using CNN-BiLSTM and Optimized SGD with Momentum
The accelerated adoption of mobile money systems has significantly increased fraudulent activity, compromising their security and trustworthiness. This research presents an enhanced method for detecting mobile money fraud by modifying a CNN-BiLSTM model with momentum using Stochastic Gradient Descent (SGD). We computed salient features from transaction data using a pre-processed hybrid CNN-BiLSTM model and trained the model to identify trends in the data that included geographical and temporal aspects. The model performed remarkably using industry-standard testing approaches: an F1 score of 0.9928, precision of 0.9927, accuracy of 0.9928, and recall of 0.9929. The proposed model can identify dishonesty and has a low false positive rate. According to the study, the model improves feature selection and incorporates various optimization techniques, making it more flexible and suitable for different mobile money systems.
- Open Access
- 24 January 2025
- Published by Springer Nature in Discover Artificial Intelligence
- Vol. 5 (1) , 1-16
AfriSign: African sign languages machine translation
- 24 January 2025
- Published by Springer Nature in Discover Artificial Intelligence
- Vol. 5 (1) , 1-16
Open AccessAfriSign: African sign languages machine translation
- Open Access
- 1 April 2024
- Published by Institute of Advanced Engineering and Science in Indonesian Journal of Electrical Engineering and Computer Science
- Vol. 34 (1) , 497-507
Sentiment analysis and classification of Ghanaian football tweets from the 2022 FIFA World Cup
Football as an attractive sport generates huge volumes of tweets concerning fans' opinions, feelings, and judgments during prime events. Such data can be leveraged in sentiment analysis, an algorithmic approach to analyzing text in tweets by extracting emotional tones. This paper presents a novel benchmark dataset of 132,115 tweets collected during the 2022 world cup on 𝕏 (formerly Twitter) for football-related sentiment classification. We also performed sentiment analysis on the dataset using lexicon-based tools, traditional machine learning algorithms, and pre-trained models, robustly optimized bidirectional encoder representations from transformers (BERT)- pretraining approach RoBERTa and distilled version of BERT (DistilBERT) to understand the emotions and reactions of football fans during different phases of the football matches. Results from the study indicate that most tweets had neutral sentiments in both context-aware and context-free analysis. We also describe our novel GhaFootBERT, a sentiment classification model based on transfer learning on BERT, which provides an effective approach to sentiment classification of football-related tweets. Our model performs robustly, outperforming the traditional models with 92% accuracy.
- 1 April 2024
- Published by Institute of Advanced Engineering and Science in Indonesian Journal of Electrical Engineering and Computer Science
- Vol. 34 (1) , 497-507
Open AccessSentiment analysis and classification of Ghanaian football tweets from the 2022 FIFA World Cup
Football as an attractive sport generates huge volumes of tweets concerning fans' opinions, feelings, and judgments during prime events. Such data can be leveraged in sentiment analysis, an algorithmic approach to analyzing text in tweets by extracting emotional tones. This paper presents a novel benchmark dataset of 132,115 tweets collected during the 2022 world cup on 𝕏 (formerly Twitter) for football-related sentiment classification. We also performed sentiment analysis on the dataset using lexicon-based tools, traditional machine learning algorithms, and pre-trained models, robustly optimized bidirectional encoder representations from transformers (BERT)- pretraining approach RoBERTa and distilled version of BERT (DistilBERT) to understand the emotions and reactions of football fans during different phases of the football matches. Results from the study indicate that most tweets had neutral sentiments in both context-aware and context-free analysis. We also describe our novel GhaFootBERT, a sentiment classification model based on transfer learning on BERT, which provides an effective approach to sentiment classification of football-related tweets. Our model performs robustly, outperforming the traditional models with 92% accuracy.