IAES International Journal of Artificial Intelligence (IJ-AI)
Not a member yet
1769 research outputs found
Sort by
Optimized pap-smear image enhancement: hybrid Perona-Malik diffusion filter-CLAHE using spider monkey optimization
Get PDFPap-smear image quality is crucial for cervical cancer detection. This study introduces an optimized hybrid approach that combines the Perona-Malik diffusion (PMD) filter with contrast-limited adaptive histogram equalization (CLAHE) to enhance pap-smear image quality. The PMD filter reduces the image noise, whereas CLAHE improves the image contrast. The hybrid method was optimized using spider monkey optimization (SMO PMD-CLAHE). Blind/reference-less image spatial quality evaluator (BRISQUE) and contrast enhancement-based image quality (CEIQ) are the new objective functions for the PMD filter and CLAHE optimization, respectively. The simulations were conducted using the SIPaKMeD dataset. The results indicate that SMO outperforms state-of-the-art methods in optimizing the PMD filter and CLAHE. The proposed method achieved an average effective measure of enhancement (EME) of 5.45, root mean square (RMS) contrast of 60.45, Michelson’s contrast (MC) of 0.995, and entropy of 6.80. This approach offers a new perspective for improving pap-smear image quality
Enhancing credit card fraud detection with synthetic minority over-sampling technique-integrated extreme learning machine
Get PDFMany works in cybersecurity detection suffer from low accuracy rates, particularly in real-world applications, where imbalanced datasets and evolving fraud strategies pose significant hurdles. This study introduces an optimized extreme learning machine (ELM) algorithm to address these challenges by dynamically adjusting hidden nodes ranging from 10 to 100 with an increment step of 10 and integrating two activation functions. The proposed method utilizes the synthetic minority over-sampling technique (SMOTE) to handle class imbalance effectively and incorporates a comprehensive evaluation using descriptive statistics, visualization, and significance testing. The proposed ELM-SMOTE method achieves the highest results including an accuracy of 99.710%, recall of 85.811%, specificity of 99.743%, and G-mean of 92.068%. These outcomes reflect the robustness and adaptability of the proposed ELM algorithm in detecting fraudulent transactions. This study emphasizes the importance of a holistic performance analysis, addressing gaps in existing methods and providing a scalable framework for real-world fraud detection applications
A bibliometric analysis of feature selection techniques: trends, innovations, and future directions
Get PDFFeature selection techniques have become increasingly important in addressing the challenges of high dimensionality in machine learning and other artificial intelligence domains. In this study, we present a comprehensive bibliometric analysis of research on feature selection techniques over the past decade, focusing on mapping the intellectual structure, identifying emerging trends, and highlighting productive collaborations in the field. Using merged data from Scopus and Web of Science databases, we collected and analyzed 2,079 relevant documents published between 2014 and 2024, applying citation analysis, co-authorship networks, and keyword co-occurrence mapping. Our findings reveal that feature selection methodologies, including supervised, unsupervised, and hybrid approaches across filter, wrapper, and embedded techniques, have been widely applied across various domains. The authors who have most contributed to the development of these methods are primarily affiliated with institutions in China, India, and the USA. The insights provided by this analysis offer researchers and practitioners a valuable foundation for guiding future research directions in feature selection
Enhancing software fault prediction through data balancing techniques and machine learning
Get PDFSoftware fault prediction is essential for ensuring the reliability and quality of software systems by identifying potential defects early in the development lifecycle. However, the presence of imbalanced datasets poses a significant challenge to the effectiveness of fault prediction models. In this paper, we investigate the impact of different data balancing techniques, including generative adversarial networks (GANs), synthetic minority over-sampling technique (SMOTE), and NearMiss, on machine learning (ML) model performance for software fault prediction. Through a comparative analysis across multiple datasets commonly used in software engineering research, we evaluate the efficacy of these techniques in addressing class imbalance and improving predictive accuracy. Our findings provide insights into the most effective approaches for handling imbalanced data in software fault prediction tasks, thereby advancing the state-of-the-art in software engineering research and practice. An extensive experimentation is performed and analyzed in this study here that includes 8 datasets, 4 data balancing techniques, and 4 ML techniques in order to demonstrate the efficacy of various models in software fault prediction
Classification of Kannada documents using novel semantic symbolic representation and selection method
Get PDFKannada is one of the 22 scheduled Indian regional languages. It is also a low-resource regional language. The Kannada document classification is arduous due to its vocabulary richness, agglutinative terms, and lack of resources. The good representation and the prominent feature selection aid in solving the challenges in document classification tasks. In this paper, we are proposing semantic symbolic representation and feature selection method, for better representation of Kannada terms in interval values embedded with positional information. Following, selection of prominent discriminative symbolic feature vectors is also proposed. Further the symbolic document classifier is used to classify the Kannada documents. The proposed cluster based symbolic representation preserves the intra class variance and reduces the ambiguity in classification of Kannada documents. The experiments are performed over two Kannada document datasets which are multilabel and unbalanced. The comparative analysis of proposed method with other standard methods is also presented
Unified BERT-LSTM framework enhances machine learning in fraud detection, financial sentiment, and biomedical classification
Get PDFThe current paper proposes a hybrid framework based on the bidirectional encoder representations from transformers (BERT) and long short-term memory (LSTM) networks for classification tasks in three diverse domains: credit card fraud detection (CCFD), financial news sentiment analysis (FNSA), and biomedical paper abstract classification (BPAC). The model leverages the strengths of BERT regarding the learning of contextual embeddings and those of LSTM in capturing sequential dependencies, thus setting the new state-of-the-art performance in each of the three domains. In the CCFD use case, the model was able to achieve an accuracy of 99.11%, considerably outperforming all the competing systems in fraud transaction detection. The BERT-LSTM model achieved a performance of 96.74% for FNSA, improving significantly in sentiment analysis. Finally, the use case of BPAC was robust, with 88.42% accuracy, which clearly classified biomedical abstract sections correctly. It is evident from the findings that this framework generalizes to a wide range of tasks and hence is an adaptable but strong tool in combating challenges of cross-domain classification
Graph based semantic email classification: a novel approach for academic institutions
Get PDFElectronic mail classification in educational institutes becomes the fundamental task to manage information efficiently. Due to the globalization and the technological advancement, volume of email users increasing consistently, which in turn increases the volume of digital data exponentially. This necessitates the developing automated email classification systems for the better and organized work. This paper develops a novel graph-based similarity (GBS) approach based on semantic similarity to address these challenges. The method initially selects the most relevant features based on feature weights, later it builds a graph by using Jaccard co efficient method for each category with features as nodes and correlation between the nodes as edges. Later, these graphs are used as templates for each category and classifies each new incoming email into the specific class based on the similarity among the graph templates and a new email. The GBS method was compared with the well-known benchmarked email classifiers and the findings demonstrated that the GBS method outperformed with 98.91% accuracy after fine-tuning of graph parameters and the classifier hyper parameters. Additionally, receiver operating characteristic (ROC) curve analysis was conducted, achieving a highest area under curve (AUC) score 0.989, demonstrating robust classification proficiency across all categories
Classifying classical music’s therapeutic effects using deep learning: a review
Get PDFMental health issues are the leading cause of global disability, increasing the need for treatment options. While there is much research on the emotional recognition of music in general, there is a gap in studies that directly connect musical features with their therapeutic effects using deep learning. This systematic literature review explores the use of deep learning in classifying the therapeutic effects of classical music for mental health. Following the preferred reporting items for systematic reviews and meta-analyses (PRISMA) framework, a total of 15 papers were reviewed. This review synthesized studies on the role of musical elements that affect mental states. Different feature extraction methods, including mel-frequency cepstral coefficients (MFCCs), spectral contrast, and chroma features, are discussed for their roles in classifying these therapeutic effects. This review also looks at deep learning algorithms like convolutional neural network (CNN), deep neural network (DNN), long short-term memory (LSTM) network, and combined models to assess their effectiveness. Common evaluation methods are also assessed to measure the performance of these models in audio classification. This review highlights the potential of combining deep learning with classical music for mental health support, and to future possibilities for applying these methods in the real world
AI-driven creativity in software development using services information
Get PDFIn competitive markets, organizations that do not innovate risk becoming outdated. At it is core, innovation depends on creativity, with creative outcomes typically characterized by novelty, usefulness, and surprisingness. In software development, creative solutions are often generated through brainstorming sessions. However, brainstorming is constrained by the knowledge of the participant and facilitator. In this paper, we present an artificial intelligence (AI)-based method to generate creative solutions in software by leveraging service information. The presented method includes two phases, where the first phase involves constructing creativity resources through text clustering (TF-IDF, K-medoid) and capability extraction (dependency parsing), and the second phase employs semantic similarity along with structured creativity techniques (exploration, transformation, and combination) to generate creative solutions in software. Besides, experimental results showed that the AI-based method achieved comparable creativity scores to traditional brainstorming with more limited time, demonstrating fast and strong performance in generating novel and useful solutions, although participants perceived some results as less surprising due to overlap with brainstorming outcomes
Hybrid intrusion detection model for hierarchical wireless sensor network using federated learning
Get PDFThe applications of wireless sensor networks are vast and popular in today’s technology world. These networks consist of small, independent sensors that are capable of measuring various physical quantities. Deployment of wireless sensor networks increased due to immense applications which are susceptible to different types of attacks in an unprotected and open region. Intrusion detection systems (IDS) play a vital part in any secured environment for any network. IDS using federated learning have the potential to achieve better classification accuracy. Usually, all the data is stored in centralized server in order to communicate between the systems. On the other hand, federated learning is a distributed learning technique that does not transfer data but trains models locally and transfers the parameters to the centralized server. The proposed research uses a hybrid IDS for wireless sensor networks using federating learning. The detection takes place in real-time through detailed analysis of attacks at different levels in a decentralized manner. Hybrid IDS are designed for node level, cluster level and the base station where federated learning acts as a client and aggregated server