IAES International Journal of Artificial Intelligence (IJ-AI)
Not a member yet
1769 research outputs found
Sort by
Generative artificial intelligence as an evaluator and feedback tool in distance learning: a case study on law implementation
Get PDFThe development of generative artificial intelligence (GAI) has impacted various fields, including higher education. This research examines the use of GAI as an evaluator and feedback provider in distance legal education. This study tested five GAI models: ChatGPT, Perplexity, Gemini, Bing, and You, using a sample of 20 students and evaluations from legal experts. Descriptive statistical analysis and non-parametric tests, including Wilcoxon, intraclass correlation coefficient (ICC), Kappa, and Kendall's W, were used to assess accuracy, feedback quality, and usability. The results showed that ChatGPT was the most effective GAI, with the highest mean scores of 4.22 from experts and 4.12 from students, followed by Gemini with scores of 4.15 and 4.07. In terms of binary judgement accuracy, Gemini scored 80%, ChatGPT 60%, while Perplexity, Bing, and You had lower scores. Statistical analysis showed moderate agreement (ICC=0.439) and low alignment (Kappa=-0.058) between the GAIs and expert evaluations, with a Kendall's W value of 0.576 indicating moderate consistency in judgements. These findings emphasize the importance of selecting effective GAIs such as ChatGPT and Gemini to improve academic evaluation and learning in legal education, and pave the way for further innovations in the use of AI
Heart disease prediction optimization using metaheuristic algorithms
Get PDFThis study explores metaheuristics hyperparameter tuning effectiveness in machine learning models for heart disease prediction. The optimized models are k-nearest neighbors (KNN) and support vector machines (SVM) using metaheuristics to identify configurations that minimize prediction error. Even though the main focus is utilizing metaheuristics to efficiently navigate the hyperparameter search space and determine optimal setting, a pre-processing and feature selection phase precedes the training phase to ensure data quality. Convergence curves and boxplots visualize the optimization process and the impact of tuning on model performance using three different metaheuristics, where an error of 0.1188 is reached. This research contributes to the field by demonstrating the potential of metaheuristics for improving heart disease prediction performance through optimized machine learning models
Lung sound classification using YAMNet, neural network, and augmentation
Get PDFGlobally, lung disease occupies a significant position as one of the main contributors to mortality rates. The characteristics of human respiratory sound signals can show a wide spectrum, ranging from normal patterns to indications of lung abnormalities. The proposed lung sound classification system is based on YAMNet as a pre-trained neural network model for medical audio recognition, which is then refined using artificial neural networks (ANN). This study presents the integration of multiple datasets and advanced pre-processing approaches. A total of 1,363 lung sound recordings from Kaggle, ICBHI, and Mendeley. This reflects the variety of clinical conditions, and differences in recording devices are combined. In order to increase the diversity of lung sound signal input, the pre-processing process is carried out through several stages, including adjusting the sampling frequency to 4 kHz, segmenting for 6 seconds, signal filtering with wavelet, min–max normalization, and data augmentation using window warping, jittering, cropping, and padding. A fold cross-validation scheme is employed to comprehensively evaluate the model's effectiveness. The evaluation results indicate that the model achieves an accuracy of 93.64%, a precision of 93.60%, a recall of 93.64%, and an F1-score of 93.52%, collectively reflecting outstanding classification performance. This work may incorporate deep learning technology into clinical practice, ultimately improving diagnosis accuracy and efficiency in the hospital setting
Revolutionizing internet of things intrusion detection using machine learning with unidirectional, bidirectional, and packet features
Get PDFDetection of attacks on internet of things (IoT) networks is an important challenge that requires effective and efficient solutions. This study proposes the use of various machine learning (ML) techniques in classifying attacks using unidirectional, bidirectional, and packet features. The proposed methods that implement decision tree (DT), random forest (RF), extreme gradient boosting classifier (XGBC), AdaBoost (AB) and linear discriminant analysis (LDA) work perfectly with all kinds of datasets and includes. It also works very well with data type-based feature selection (DTBFS) and correlation-based feature selection (CBFS). The experiment results show a significant improvement compared to previous studies and reveals that unidirectional and bidirectional features provide higher accuracy compared to packet features. Furthermore, ML models, particularly DT, and RF, have faster computing times compared to more complex deep learning models. This analysis also shows potential overfitting in some models, which requires further validation with different datasets. Based on these findings, we recommend the use of RF and DT for scenarios with unidirectional and bidirectional features, while AB and LDA for packet features. The study concludes that using the right ML techniques along with features that work in both directions can make an intrusion detection system for IoT networks becomes very accurate
Hybridized deep learning model with novel recommender for predicting criticality state of patient using MIMIC-IV dataset
Get PDFThe contribution of machine learning towards prediction of critical state of patient is the prime focus of the current study. The review of current approaches of machine learning has been witnessed with various shortcomings. Hence, the proposed study adopts medical information mart for intensive care (MIMIC-IV) dataset in order to develop a novel analytical model that can predict the criticality state of patient in their next visit. The model has been designed by hybridizing convolution neural network (CNN) and long short-term memory (LSTM) which takes the discrete input of hospital and individual patient information in each visit. The concatenated feature is then subjected to a newly introduced recommender module which offers implicit feedback by assigning a ranking score. The final predictive outcome of study offers criticality rank. The study model is benchmarked with existing machine learning approaches to find 54% of increased accuracy and 70% of reduced processing time
A fusion convolution neural network-local binary pattern histogram algorithm for emotion recognition in human
Get PDFThis paper proposes a fusion of algorithms namely convolution neural networks (CNN) and local binary pattern histogram (LBPH) techniques to comprehend the emotions in humans for greyscale images. In this work, the combined advantages of CNN for its ability to extract features, suitability for image processing and LBPH algorithm to identify the emotions of the human images are included. Though there are enhanced fused algorithms with CNN for image processing, the combination of LBPH with CNN is precise and simple in design. In this work, the secondary data sample is used to recognize the human emotions. The secondary data set consists of 160 samples with emotions of happy, anger, sad, and surprise is considered for making decisions. In comparison, the accuracy of the proposed method is high compared to the other algorithms
Dual simulated annealing soft decoder for linear block codes
Get PDFThis paper proposes a new approach to soft decoding for linear block codes called dual simulated annealing soft decoder (DSASD) which utilizes the dual code instead of the original code, using the simulated annealing algorithm as presented in a previously developed work. The DSASD algorithm demonstrates superior decoding performance across a wide range of codes, outperforming classical simulated annealing and several other tested decoders. We conduct a comprehensive evaluation of the proposed algorithm's performance, optimizing its parameters to achieve the best possible results. Additionally, we compare its decoding performance and algorithmic complexity with other decoding algorithms in its category. Our results demonstrate a gain in performance of approximately 2.5 dB at a bit error rate (BER) of 6×10⁻⁶ for the LDPC (60,30) code
Strid-CNN: moving filters with convolution neural network for multi-class pneumonia classification
Get PDFMillions of people around the world suffer from pneumonia, a serious lung illness. To effectively treat and manage this condition, a quick and accurate diagnosis is essential. This study thoroughly examines different ways of using transfer learning to classify pneumonia into multiple categories. We use well-known methods like DenseNet121, VGGNet-16, ResNet-50, and Inception Net, as well as a new method called Strid-CNN, which applies moving filters with convolution neural network. Through extensive testing, we show that each method effectively uses pre-learned information on a large dataset of medical images, accurately identifying pneumonia across various classes. Our results reveal subtle differences in performance among these methods, providing insights into how well they adapt to the challenging field of medical image analysis. Additionally, the Strid-CNN method shows promising results, indicating its potential as a competitive alternative. This research offers valuable guidance on choosing the right transfer learning approach for classifying pneumonia into multiple categories, contributing to improvements in diagnostic accuracy and healthcare effectiveness. Our study not only highlights the current state of transfer learning in pneumonia classification but also its potential to enhance clinical outcomes and patient care
A novel scalable deep ensemble learning framework for big data classification via MapReduce integration
Get PDFBig data classification involves the systematic sorting and analysis of extensive datasets that are aggregated from a variety of sources. These datasets may include but are not limited to, electronic records, digital imaging, genetic information sequences, transactional data, research outputs, and data streams from wearable technologies and connected devices. This paper introduces the scalable deep ensemble learning framework for big data classification (SDELF-BDC), a novel methodology tailored for the classification of large-scale data. At its core, SDELF-BDC leverages a Hadoop-based map-reduce framework for feature selection, significantly reducing feature-length and enhancing computational efficiency. The methodology is further augmented by a deep ensemble model that judiciously applies a variety of deep learning classifiers based on data characteristics, thereby ensuring optimal performance. Each classifier's output undergoes a rigorous optimization-based ensemble approach for refinement, utilizing a sophisticated algorithm. The result is a robust classification system that excels in predictive accuracy while maintaining scalability and responsiveness to the dynamic requirements of big data environments. Through a strategic combination of classifiers and an innovative reduction phase, SDELF-BDC emerges as a comprehensive solution for big data classification challenges, setting new benchmarks for predictive analytics in diverse and data-intensive domains
Advancing integrity and privacy in cloud storage: challenges, current solutions, and future directions
Get PDFThe rapid expansion of cloud computing has steered in an era where cloud storage is increasingly prevalent, offering significant advantages in terms of reducing local storage burden. However, this technological shift has also introduced complex security challenges, including data integrity and privacy concerns. In response to these challenges, various data integrity auditing (DIA) protocols have been developed, aiming to enable efficient and secure verification of data stored in cloud environments. This survey paper provides a comprehensive analysis of existing DIA mechanisms, focusing on methods like homomorphic linear authentication, dynamic hash tables, and watermarking techniques for integrity and privacy preservation. It critically evaluates these methods in terms of their advantages, limitations, and the unique challenges they face in practical applications, such as scalability, efficiency in multi-owner contexts, and real-time auditing. Furthermore, the paper identifies key research gaps, including the need for optimizing largescale data handling, balancing watermarking imperceptibility with embedding capacity, and developing comprehensive solutions for decentralized public auditing. The survey serves as a critical resource for researchers to understand the current background of cloud data integrity auditing and the future directions in this evolving field