Communities in DSpace

Select a community to browse its collections.

Recent Submissions

Thumbnail Image
Item
Anomaly Detection in Social Media Conversation Using Natural Language Processing with LSTM and Naïve Bayes Models
(Journal of The Faculty of Computational Sciences & Informatics Academic City University College Accra, Ghana, 2024) Babatunde A.N.,Shuaib B.M.,Kadri A.F., Isiaka O.S.,Abdulrahman, T.A., Ismail, S.I. & Oke A.A
While these platforms are now used widely, they have also resulted in the rise of problematic or discursive and offensive material which has serious implications for online safety. To tackle this issue, this paper presents a hybrid method combining Natural Language Processing (NLP) for data preprocessing, Long-Short-Term-Memory (LSTM) with Naïve Bayes (NB) model to detect anomalies in the conversational style of users on social media. The NLP was used to process raw tweet data which was taken from the Kaggle dataset containing 24,783 instances with 6 features and processed to a format that is ideal for analysis. Naïve Bayes and LSTM models were then trained to detect abnormal or problematic messages such as hate speech and offensive language after data cleaning and transformation. Naïve Bayes, which served as a fast, probabilistic baseline model for the classification of the text was complemented in this research with the ability of the LSTM model to provide deep contextual understanding as well as sequential patterns that occur in the conversation. The model’s performance metrics followed standard Machine Learning practices, and the best result was presented by the hybrid LSTM-NB model with an accuracy of 99.2% while the NB had 90% test result and LSTM 95% test score. This impressive result highlights the benefits of deploying NLP, integrated with both traditional machine learning and deep learning techniques to deal with the issue of detecting anomalies on online conversations. This paper aims to add to the NLP and AI literature, by offering an economic model to make digital communication safer.
Thumbnail Image
Item
Sentiments Analysis of Social Media Text Using XGboost, LightGBM, and Fasttext Embeddings for Opinion Mining
(Published by: Faculty of Computing and Information Technology and Faculty of Natural and Applied Science, Sule lamido University, Kafin Hausa, Jigawa State, Nigeria. Available online at Sule Lamido University - Journal of Science and Technology., 2025) Mohammed, S. B., Kadri, A. F., Abdulrahman T. A., Ilori, O., Tajudeen, K. O., Sulaiman, H. O., Ekundayo, O., & Babatunde A.N
Sentiment analysis is an important task in Natural Language Processing (NLP), aiming to classify text into categories such as positive, negative, or neutral. Despite considerable progress, existing models often struggle with the noisy, short-text data that characterizes social media platforms. Aim: This study proposes a sentiment classification framework built on the Sentiment140 dataset, where preprocessing steps and FastText embeddings were used to generate dense vector representations of tweets. Positive, negative, and a derived neutral class were defined as target categories. Method: Two gradient boosting algorithms, XGBoost and LightGBM, were trained and evaluated across five independent runs using standard metrics, including accuracy, precision, recall, and F1-score. Results: The results demonstrate that XGBoost consistently achieved superior performance, with an average accuracy of 97.8% ± 0.4 and all other metrics exceeding 97%, while LightGBM recorded substantially lower performance with an average accuracy of 75.2% ± 1.8. Baseline comparisons with majority-class prediction, Logistic Regression, and SVM confirmed the robustness of the proposed framework, while benchmarking against previously reported CNN and LSTM models showed a clear improvement over existing state-of-the-art result. These findings validate the effectiveness of combining FastText embeddings with XGBoost for sentiment classification and emphasize its reproducibility and robustness. The study provides methodological clarity, addresses the limitations of prior approaches, and establishes a reliable foundation for real-time applications in social media analytics, customer experience monitoring, and decision support systems.
Thumbnail Image
Item
Application of Residue Number System in Information Security: A Systematic Review
(Sule Lamido University Journal of Science and Technology, 2025) Babatunde, A.N., Balogun, B.F., Lawal A.O., Akuji, I.I., Kolawole O.F., Isiaka, O.S., Oke, A.A., Kadri, A.F., Shuaib, B.M., Sulaiman, H.B. & Elelu, A.T
The Residue Number System (RNS) offers a promising alternative to traditional weighted arithmetic systems, demonstrating potential advantages in cryptography, error detection, and data integrity. However, its practical adoption remains limited due to challenges in base conversion, scalability, and hardware efficiency. Aim: This study conducts a systematic literature review (SLR) to evaluate the role of RNS in securing sensitive data, with a focus on its computational efficiency, fault tolerance, and security enhancements. Methodology: A comprehensive review of 100 high-quality studies published between 2017 and 2024 was conducted, analyzing RNS's impact on cryptographic algorithms, error detection mechanisms, and data integrity protocols. The study also identifies existing research gaps and assesses the feasibility of integrating RNS into real-world security frameworks. Results: The findings indicate that RNS significantly improves the performance of cryptographic systems such as RSA, AES, and elliptic curve cryptography (ECC), reducing computational complexity and enhancing processing speed by 20-50%. Additionally, RNS enhances fault tolerance and error detection, making it a viable approach for secure data transmission and storage. However, practical implementation remains challenging due to the computational overhead of base conversion, hardware constraints, and integration difficulties in large-scale cryptographic frameworks. RNS holds significant promise for enhancing information security, but addressing its practical limitations is crucial for broader adoption. Future research should focus on optimizing base conversion techniques, reducing hardware overhead, and exploring hybrid cryptographic models to improve real-world feasibility.
Thumbnail Image
Item
Evaluation of the least significant bit-based steganography’s performance using elliptic curve cryptography
(Kasu Journal of Computer Science, 2024) Tajudeen, K. O., Kadri, A. F., & Aro, O. T
The importance of data security is growing for companies trying to safeguard sensitive information. Cryptography and steganography are two widely used methods. Steganography enables the concealment of secret information within digital content to reduce suspicions. Even though the Least Significant Bit (LSB) technique is widely utilized, when applied alone, it is still vulnerable to security breaches. Traditional approaches that combine LSB steganography with RSA encryption enhance security but are computationally expensive, making them less useful for real-time applications like data storage, secure communication, and the Internet of Things. The need for a robust yet lightweight system that combines encryption and steganography without compromising performance is thus necessary. This study proposes a hybrid approach that blends LSB steganography with Elliptic Curve Cryptography (ECC). Because it uses less processing power and has lower key sizes than RSA, ECC was selected to maintain similar security standards. The method was implemented and evaluated in terms of encryption time, data-hiding capabilities, and efficiency. Experimental results show that the LSB–ECC framework outperforms the conventional LSB–RSA combination. Specifically, by significantly reducing encryption time and improving hiding efficiency, ECC offers a more scalable solution for safeguarding sensitive data. The findings confirm that ECC provides a lighter solution without sacrificing security. ECC's integration with LSB steganography exhibits exceptional efficiency and security performance, making it appropriate for contemporary data protection requirements. Adaptive steganographic approaches and testing on extensive multimedia datasets will be incorporated into future work to further validate robustness against sophisticated steganalysis and cryptographic attacks.
Thumbnail Image
Item
Machine Learning for Carbon Emissions Prediction and Anomaly Detection: A Comparative Study of Traditional and Deep Learning Approaches
(KWASU Journal of Information, Communication and Technology (KJICT.), 2026) Rafiu Mope ISIAKA , Akeem Femi KADRI , Oluwasayo EKUNDAYO ,Habeeb Olayinka SULAIMAN
Accurate greenhouse gas emissions monitoring is critical for evaluating climate policies and ensuring regulatory compliance. Although machine learning offers promising capabilities, systematic comparisons between tradition al and deep learning approaches for structured environmental time-series data remain limited. Here, we evaluated seven prediction models and three anomaly detection approaches on monthly CO₂ emissions data from 20 major emitting countries between 2000 and 2024. Random Forest achieved superior performance (R² = 0.902) com pared to all deep learning architectures (R² ≈ 0), while requiring 15–60 times less training time. SHapley Additive exPlanations revealed that the 12-month lag feature dominated the predictions (75.6% importance), indicating strong year-over-year persistence with direct policy implications. Deep learning failure stems from insufficient training data (3,863 samples), structured tabular characteristics with domain-engineered features, and unnecessary architectural complexity, thereby aligning with emerging evidence that traditional methods often outperform deep learning on tabular datasets. Anomaly detection proved challenging, with all unsupervised models achieving F1 scores below 0.05, suggesting that supervised approaches with labelled examples are necessary for operational deployment. We recommend prioritising Random Forest for prediction, investing in domain-informed feature en gineering, integrating explainability tools for transparency, and deploying anomaly detection with human review. We identified eight research frontiers, including satellite data integration, federated learning, causal inference, and hybrid physics-machine learning models. This study demonstrates that traditional machine learning, combined with domain expertise and interpretability tools, can provide accurate, efficient, and transparent emission predic tions for operational climate monitoring systems.