IAES International Journal of Artificial Intelligence (IJ-AI)
Not a member yet
1769 research outputs found
Sort by
Predicting enhanced diagnostic models: deep learning for multi-label retinal disease classification
Get PDFIn this study, we assess three convolutional neural network (CNN) architectures—VGG16, ResNet50, and InceptionV3 for multi classification of fundus images in the retinal fundus multi-disease image dataset (RFMID2), comprising of 860 images. Focusing on diabetic retinopathy, exudation, and hemorrhagic retinopathy, we preprocessed the dataset for uniformity and balance. Using transfer learning, the models were adapted for feature extraction and fine-tuned to our multi-label classification task. Their performance was measured by subset accuracy, precision, recall, F1-score, hamming loss, and Jaccard score. VGG16 emerged as the top performer, with the highest subset accuracy (84.81%) and macro precision (95.83%), indicating its superior class distinction capabilities. ResNet50 showed commendable accuracy (79.75%) and precision (86.70%), whereas InceptionV3 lagged with lower accuracy (66.67%) and precision (81.21%). These findings suggest VGG16’s depth offers advantages in multi-label classification, highlighting InceptionV3’s limitations in complex scenarios. This analysis helps optimize CNN architecture selection for specific tasks, suggesting future exploration of dataset variability, ensemble methods, and hybrid models for improved performance
Reliable backdoor attack detection for various size of backdoor triggers
Get PDFBackdoor attack techniques have evolved toward compromising the integrity of deep learning (DL) models. To defend against backdoor attacks, neural cleanse (NC) has been proposed as a promising backdoor attack detection method. NC detects the existence of a backdoor trigger by inserting perturbation into a benign image and then capturing the abnormality of inserted perturbation. However, NC has a significant limitation such that it fails to detect a backdoor trigger when its size exceeds a certain threshold that can be measured in anomaly index (AI). To overcome such limitation, in this paper, we propose a reliable backdoor attack detection method that successfully detects backdoor attacks regardless of the backdoor trigger size. Specifically, our proposed method inserts perturbation to backdoor images to induce them to be classified into different labels and measures the abnormality of perturbation. Thus, we assume that the amount of perturbation required to reclassify the label of backdoor images to the ground-truth label will be abnormally small compared to them for other labels. By implementing and conducting comparative experiments, we confirmed that our idea is valid, and our proposed method outperforms an existing backdoor detection method (NC) by 30%p on average in terms of backdoor detection accuracy (BDA)
Optimizing seismic sequence clustering with rapid cube-based spatiotemporal approach
Get PDFDue to their extensive volume and range of features, seismic data is regarded as highly complex data. Earthquakes that typically composed of foreshocks, mainshocks, and aftershocks, exhibit a unique sensitivity to temporal dimension, a characteristic that differs them from other natural hazards. Foreshocks and aftershocks that emanate from a similar epicenter, often display temporal patterns that contribute significantly to determining a sequence. This study introduces a density cube-based approach to cluster spatiotemporal seismic data. It addresses spatial irregularities observed in earthquake clusters and incorporates temporal aspects, acknowledging that seismic events originating from a similar epicenter could occur in separate time frames. We achieved the highest Silhouette score of 0.935 in daily-based clustering and 0.782 in weekly-based clustering. Notably, our analysis reveals a trend where weekly clustering lambda λ tend to be lower (λ=0.01) than in daily clustering (λ=0.1, λ=0.5), thus emphasizing the significance of temporal granularity where daily clustering requires higher λ to capture rapid fluctuations, while weekly clustering benefits from lower λ to cover broader trends. These findings enhance the understanding of the nuanced interplay of temporal dynamics in seismic sequence analysis
Chinese paper classification based on pre-trained language model and hybrid deep learning method
Get PDFWith the explosive growth in the number of published papers, researchers must filter papers by category to improve retrieval efficiency. The features of data can be learned through complex network structures of deep learning models without the need for manual definition and extraction in advance, resulting in better processing performance for large datasets. In our study, the pre-trained language model bidirectional encoder representations from transformers (BERT) and other deep learning models were applied to paper classification. A large-scale chinese scientific literature dataset was used, including abstracts, keywords, titles, disciplines, and categories from 396 k papers. Currently, there is little in-depth research on the role of titles, abstracts, and keywords in classification and how they are used in combination. To address this issue, we evaluated classification results by employing different title, abstract, and keywords concatenation methods to generate model input data, and compared the effects of a single sentence or sentence pair data input methods. We also adopted an ensemble learning approach to integrate the results of models that processed titles, keywords, and abstracts independently to find the best combination. Finally, we studied the combination of different types of models, such as the combination of BERT and convolutional neural networks (CNN), and measured the performance by accuracy, weighted average precision, weighted average recall, and weighted average F1 score
Primary phase Alzheimer's disease detection using ensemble learning model
Get PDFAlzheimer's disease (AD) is a noteworthy problem for public health. Older people are most impacted by this neurological disease. It leads to memory loss and various cognitive impairments, eventually hindering communication. As a result, research on early AD detection has intensified in recent years. In current research work, we propose an ensemble learning strategy to identify AD by classifying brain images into two groups: AD brain and normal brain. Researchers have recently explored various machine learning (ML) and deep learning techniques to improve early disease detection. Patients with AD can recover from it more successfully and with less damage if they receive early diagnosis and therapy. This research presents an ensemble learning model to predict AD using decision trees (DT), logistic regression (LR), support vector machines (SVM), and convolutional neural networks (CNN). The open access series of imaging studies (OASIS) dataset is used for model training, and performance is measured in terms of various kinds of outcome namely accuracy, precision, recall, and F1 score. Our results demonstrated that, for the AD dataset, the CNN achieved the maximum validation accuracy of 90.32%. Thus, by accurately detecting the condition, ensemble algorithms can potentially significantly reduce the annual mortality rates associated with AD
Predicting tourist arrivals to a tropical island using artificial intelligence
Get PDFThis research leverages artificial intelligence (AI) techniques to develop a predictive model for forecasting tourist arrivals in East Java Province, Indonesia, using a comprehensive dataset encompassing historical tourism statistics from 2018 to 2020, seasonal trends, promotional campaigns, and various economic and social variables. The study evaluates three AI methodologies: artificial neural network (ANN), extreme learning machine (ELM), and Jordan recurrent neural network (JRNN), each known for their distinct strengths in processing complex data and adapting to changing trends. The comparative analysis reveals that the JRNN model outperforms others with the highest precision, achieving an average prediction deviation of just 2.98% from actual data, effectively capturing temporal and seasonal trends. The ANN follows closely with a deviation of 3.31%, showing strong capabilities in handling complex, nonlinear relationships. In contrast, the ELM, though fastest in training, exhibits a larger deviation of 10.51%, indicating a trade-off between speed and accuracy. These results highlight the potential of AI to significantly enhance the accuracy and operational efficiency of tourism forecasts, offering robust tools for stakeholders to engage in informed strategic planning and resource allocation in dynamic market conditions
Two-dimensional Klein-Gordon and Sine-Gordon numerical solutions based on deep neural network
Get PDFDue to the well-known dimensionality curse, developing effective numerical techniques to resolve partial differential equations proved a complex problem. We propose a deep learning technique for solving these problems. Feedforward neural networks (FNNs) use to approximate a partial differential equation with more robust and weaker boundaries and initial conditions. The framework called PyDEns could handle calculation fields that are not regular. Numerical exper- iments on two-dimensional Sine-Gordon and Klein-Gordon systems show the provided frameworks to be sufficiently accurate
A comparative analysis of exponential smoothing method and deep learning models for bitcoin price prediction
Get PDFBlockchain technology is the foundation of cryptocurrencies, which are virtual currencies. The decentralized nature of cryptocurrencies has resulted in a significant reduction of central authority over them, which has implications for global trade and relations. The need for an effective model to anticipate the price of cryptocurrencies is essential due to their wide variations in value. Due to the shortcomings of conventional production forecasting, in this work, four distinct models were used. The deep learning models are the long short-term memory (LSTM) and bidirectional long short-term memory (Bi-LSTM), and both the Facebook-Prophet and Silverkite support the exponential smoothing technique. Silverkite is designed to handle a wide range of time series forecasting tasks. Considering past bitcoin information from January 2012 to March 2021, a period of nine years, we looked at the models. The Bi-LSTM model yields a 7.073 mean absolute error (MAE) and a 3.639 root mean squared error (RMSE). The Bi-LSTM model identifies the deviations that might draw attention and avert any problems
Enhanced intrusion detection through dual reduction and robust mean
Get PDFThe exponential growth of online networks necessitates a paradigm shift in intrusion detection systems (IDS). Traditionalmethods falter under the massive influx of data, resulting in high false positives and reduced detection accuracy. This research introduces a novel approach combining principal component analysis (PCA) and linear discriminant analysis (LDA), augmented by robust generalized sample mean, to enhance IDS performance. PCA efficiently reduces data dimensionality, while LDA extracts critical features that differentiate normal network traffic from anomalies. The robust generalized sample mean counteracts the influence of outliers, ensuring accurate and reliable analysis. Implemented on the UNSW-NB15 dataset, our method achieves an average 6% reduction in false positives and a 10% increase in detection rate. Additionally, our testing methodology closely mirrors real-world conditions, making the results more representative of practical scenarios compared to existing work. These advancements demonstrate substantial improvements in IDS performance and robustness over existing techniques
Development of a 2 degree of freedom-proportional integral derivative controller using the hippopotamus algorithm
Get PDFThis research project investigates the regulation of autonomous power generation in two interconnected regions using two hydroelectric power plants. It specifically addresses the challenges posed by significant electrical system issues. The hippopotamus optimization algorithm (HOA) has demonstrated enhanced gain value in research and designs of 2 degree of freedom (2DOF)-proportional integral derivative (PID) controllers. The objective is to provide efficient and uninterrupted functioning of the electrical network in both areas. Contemporary technology and methods enable the electrical system to efficiently and accurately fulfill user requirements, resolving any problems related to system balance and stability. This experiment evaluates the efficacy of several algorithms in accurately selecting optimal values. We evaluate performance using the integral of absolute error (IAE) and integral of time-weighted absolute error (ITAE) functions. This experiment evaluates and contrasts different algorithms. Summarizing the analysis using verifiable evidence. Optimization when evaluated using the ITAE measurement, the HOA earned the lowest result of 0.08744 for ITAE. Empirical research has demonstrated that this strategy is the most effective in reducing the ITAE. The sine-cosine algorithm (SCA) and whale optimization algorithm (WOA) have similar ITAE values, with SCA having an error of 0.08967 and WOA having an error of 0.08967. The numerical number is 0.08970