IAES International Journal of Artificial Intelligence (IJ-AI)
Not a member yet
1769 research outputs found
Sort by
Learning high-level spectral-spatial features for hyperspectral image classification with insufficient labeled samples
Get PDFHyperspectral image (HSI) classification research is a hot area, with a mass of new methods being developed to improve performance for specific applications that use spatial and spectral image material. However, the main obstacle for scientists is determining how to identify HSIs effectively. These obstacles include an increased presence of redundant spectral information, high dimensionality in observed data, and limited spatial features in a classification model. To this end, we, therefore, proposed a novel approach for learning high-level spectral-spatial features for HSI classification with insufficient labeled samples. First, we implemented the principal component analysis (PCA) technique to reduce the high dimensionalities experienced. Second, a fusion of 2D and 3D convolutions and DenseNet, a transfer learning network for feature learning of both spatial-spectral pixels. The achieved experimental results are comparatively satisfactory to contrasted approaches on the widely used HSI images, i.e., the University of Pavia and Indian Pines, with an overall classification accuracy of 97.80% and 97.60%, respectively
Enhancing facial recognition accuracy through feature extractions and artificial neural networks
Get PDFFacial recognition is a biometric system used to identify individuals through faces. Although this technology has many advantages, it still faces several challenges. One of the main challenges is that the level of accuracy has yet to reach its maximum potential. This research aims to improve facial recognition performance by applying the discrete cosine transform (DCT) and Gaussian mixture model (GMM), which are then trained with backward propagation of errors (backpropagation) and convolutional neural networks (CNN). The research results show low DCT and GMM feature extraction accuracy with backpropagation of 4.88%. However, the combination of DCT, GMM, and CNN feature extraction produces an accuracy of up to 98.2% and a training time of 360 seconds on the Olivetti Research Laboratory (ORL) dataset, an accuracy of 98.9% and a training time of 1210 seconds on the Yale dataset, and 100% accuracy and training time 1749 seconds on the Japanese female facial expression (JAFFE) dataset. This improvement is due to the combination of DCT, GMM, and CNN's ability to remove noise and study images accurately. This research is expected to significantly contribute to overcoming accuracy challenges and increasing the flexibility of facial recognition systems in various practical situations, as well as the potential to improve security and reliability in security and biometrics
Explainable machine learning models applied to predicting customer churn for e-commerce
Get PDFPrecise identification of customer churn is crucial for e-commerce companies due to the high costs associated with acquiring new customers. In this sector, where revenues are affected by customer churn, the challenge is intensified by the diversity of product choices offered on various marketplaces. Customers can easily switch from one platform to another, emphasizing the need for accurate churn classification to anticipate revenue fluctuations in e-commerce. In this context, this study proposes seven machine learning classification models to predict customer churn, including decision tree (DT), random forest (RF), support vector machine (SVM), logistic regression (LR), naïve Bayes (NB), k-nearest neighbors (K-NN), and artificial neural network (ANN). The performances of the models were evaluated using confusion matrix, accuracy, precision, recall, and F1-score. The results indicated that the ANN model achieves the highest accuracy at 92.09%, closely followed by RF at 91.21%. In contrast, the NB model performed the least favorably with an accuracy of 75.04%. Two explainable artificial intelligence (XAI) methods, shapley additive explanations (SHAP) and local interpretable model-agnostic explanations (LIME), were used to explain the models. SHAP provided global explanations for both ANN and RF models through Kernel SHAP and Tree SHAP. LIME, offering local explanations, was applied only to the ANN model which gave better accuracy
Adaptive kernel integration in visual geometry group 16 for enhanced classification of diabetic retinopathy stages in retinal images
Get PDFDiabetic retinopathy (DR) is a major cause of vision impairment globally, with early detection remaining a significant challenge. The limitations of current diagnostic methods, particularly in identifying early-stage DR, highlight a pressing need for more accurate diagnostic technologies. In response, our research introduces an innovative model that enhances the visual geometry group 16 (VGG16) architecture with adaptive kernel techniques. Traditionally, the VGG16 model deploys consistent kernel sizes throughout its convolutional layers. In this study, multiple convolutional branches with varying kernel sizes (3×3, 5×5, and 7×7) were seamlessly integrated after the 'block5_conv1' layer of VGG16. These branches were adaptively merged using a softmax-weighted combination, enabling the model to automatically prioritize kernel sizes based on the image's intricate features. To combat the challenge of imbalanced datasets, the synthetic minority over-sampling technique (SMOTE) was employed before training, harmonizing the distribution of the five DR stages. Our results are promising, showing a training accuracy above 94.17% and a validation accuracy over 90.24%, our model significantly outperforms traditional methods. This study represents a significant stride in applying adaptive kernels to deep learning for precise medical imaging tasks. The model's accuracy in classifying DR stages highlights its potential as a valuable diagnostic tool, paving the way for future enhancements in DR detection and management
A portfolio optimization model for return trend rate and risk trend rate based on machine learning
Get PDFThis paper presents a machine learning-based portfolio optimization model alongside a trading strategy algorithm. There are two distinct steps to the approach. Firstly, the long short-term memory (LSTM) neural network model was used to predict the closing price of stocks in the following 4 days. The average rise and fall rate over these four days is then calculated as the stock's return trend rate, which can measure the direction and intensity of the stock's rise and fall. The same method is used to predict the average of the industry index's rise and fall rate over the next four days as the risk trend rate. In the second step, the improved mean–variance model (IMV) model is used to provide customers with the stock portfolio purchasing strategy based on the return trend rate and risk trend rate. The experimental results demonstrate that the approach has a certain application value and outperforms the traditional method in terms of annual returns and Sharpe ratio, using the Shanghai Stock Exchange and the Shenzhen Stock Exchange as study samples. The model shows approximately 1% improvement in prediction accuracy. The latest advancements in machine learning provide substantial prospects for tactics involving the purchase of portfolios
A robust penalty regression function-based deep convolutional neural network for accurate cardiac arrhythmia classification using electrocardiogram signals
Get PDFCardiac arrhythmias are a leading cause of morbidity and mortality worldwide, necessitating accurate, and timely diagnosis. This paper presents a novel approach for the classification of cardiac arrhythmias using a penalty regression function (PRF)-based deep convolutional neural network (DCNN). The proposed model integrates advanced preprocessing techniques, including frechet with fitness rank distribution-based anas platyrhynchos optimization (FFRD-APO) for feature selection and ensemble empirical mode decomposition (EEMD) for signal decomposition. Utilizing the St. Petersburg INCART 12-lead arrhythmia database, the PRF-DCNN model achieved superior performance metrics: an area under the curve-receiver operating characteristic (AUC-ROC) of 0.97, accuracy of 0.95, precision of 0.93, recall of 0.92, specificity of 0.97, and an F1 score of 0.93. The PRF effectively mitigated overfitting, ensuring robust and reliable classification across varied patient demographics. The model demonstrated significant improvements over traditional methods, offering an efficient solution for real-time cardiac monitoring and diagnosis. This study underscores the potential of PRF-DCNN in enhancing automated arrhythmia detection and lays the groundwork for future research to optimize and validate this approach in diverse clinical settings
Efficient reduction of computational complexity in video surveillance using hybrid machine learning for event recognition
Get PDFThis paper addresses the challenge of high computational complexity in video surveillance systems by proposing an efficient model that integrates hybrid machine learning algorithms (HML) for event recognition. Conventional surveillance methods struggle with processing vast amounts of video data in real-time, leading to scalability, and performance issues. Our proposed approach utilizes convolutional neural networks (CNNs) and recurrent neural networks (RNNs) to enhance the accuracy and efficiency of detecting events. By comparing our model with conventional surveillance techniques motion detection, background subtraction, and frame differencing. We demonstrate significant improvements in frame processing time, object detection speed, energy efficiency, and anomaly detection accuracy. The integration of dynamic model scaling and edge computing further optimizes computational resource usage, making our method a scalable and effective solution for real-time surveillance needs. This research highlights the potential of machine learning to revolutionize video surveillance, offering insights into developing more intelligent and responsive security systems. The results of your simulation analysis, indicating performance improvements in accuracy by 0.25%, 0.35%, and 0.45% for the motion detection algorithm, background subtraction, and frame differencing respectively, and in real-time data processing by 5.65%, 4.45%, and 6.75% for the motion detection algorithm, background subtraction, and frame differencing respectively, highlight the potential of machine learning to transform video surveillance into a more intelligent and responsive system
Deep learning-based techniques for video enhancement, compression and restoration
Get PDFVideo processing is essential in entertainment, surveillance, and communication. This research presents a strong framework that improves video clarity and decreases bitrate via advanced restoration and compression methods. The suggested framework merges various deep learning models such as super-resolution, deblurring, denoising, and frame interpolation, in addition to a competent compression model. Video frames are first compressed using the libx265 codec in order to reduce bitrate and storage needs. After compression, restoration techniques deal with issues like noise, blur, and loss of detail. The video restoration transformer (VRT) uses deep learning to greatly enhance video quality by reducing compression artifacts. The frame resolution is improved by the super-resolution model, motion blur is fixed by the deblurring model, and noise is reduced by the denoising model, resulting in clearer frames. Frame interpolation creates additional frames between existing frames to create a smoother video viewing experience. Experimental findings show that this system successfully improves video quality and decreases artifacts, providing better perceptual quality and fidelity. The real-time processing capabilities of the technology make it well-suited for use in video streaming, surveillance, and digital cinema
A multi-algorithm approach for phishing uniform resource locator’s detection
Get PDFNowadays, the internet is used to organise a wide range of cybersecurity risks. Threats to cybersecurity include a broad spectrum of malevolent actions and possible hazards that affect data, networks, and digital systems. Cybersecurity dangers that are commonly encountered are distributed denial-of-service (DDoS) attacks, phishing, and malware. Phishing attempts frequently use text messages, email, and uniform resource locators (URLs) to target specific people while impersonating trustworthy sourcesin an effort to trick the victim. Consequently, machine learning plays a critical role in stopping cybercrimes, especially those that involve phishing assaults. The suggested model is based on a well constructed dataset that has been enhanced with 32 features. By combining the features of several machine learning methods, such as random forest, CatBoost, AdaBoost, and multilayer perceptron, the suggested model greatly increases the precision of phishing URL detection. Evaluation indicators that highlight the model's effectiveness in defending against cyber threats include precision, recall, accuracy, and F1-score. These metrics also highlight the urgent need for proactive cybersecurity measures
A mixed integer nonlinear programming model for site-specific management zone problem
Get PDFPrecision agriculture employs sophisticated tools to optimize decision-making in farming, aiming to simultaneously improve crop yields and manage resources more effectively in a context of increasing scarcity and rising costs. A key aspect of precision agriculture is the delineation of site-specific management zones (SSMZs), which involves segmenting a field into areas that are homogeneous in terms of soil physicochemical properties. The problem of delineating SSMZ have been approached using a wide variety of methodologies, all of which, heuristic, focus on finding feasible solutions. Until this work, there was no exact algorithm or mathematical model that would allow for a point of comparison. This paper introduces a novel approach to tackle the delineation of SSMZ with orthogonal shapes through the development of a mixed integer nonlinear programming (MINLP) model. Small instances with different scenarios show the scope of the proposed approach and the significance of the results. It provides a structure for the SSMZ problem with orthogonal shapes and establishes a benchmark for evaluating the performance of heuristic solutions, metaheuristics, or hybrid approaches