PURPOSE: The earlier prediction of autism spectrum disorder (ASD) placed a serious attention on ensuring the appropriate intervention to improve the child's behavioral, cognitive, and social development. The previous detection process is commonly time-intensive, subjective, and highly dependent on the clinical professions, which leads to limited accessibility in rural areas. The difficulties are addressed by introducing effective ASD detection systems, which provide a scalable, objective, and fast solution, reducing the challenges in the healthcare environment. METHOD: This work integrates the Glowworm Optimization with Extreme Learning Machine Networks (GO-ELMN) model to enhance the efficiency of ASD prediction. During the analysis, ASD screening data for children are collected and processed frequently to obtain behavioral, demographic, and medical features. The extracted features are processed by an extraction learning technique, in which the network hyperparameters are optimized using the glowworm optimization algorithm. The optimized classifier recognizes children's behavior by addressing the issues of limited and imbalanced data. FINDINGS: The efficiency of the system is evaluated using experimental results, in which the system ensures high accuracy and convergence speed. CONCLUSION: The ASD detection model provides an interpretable, fast, and reliable solution that is effectively utilized in the pediatric healthcare domain.
Diabetes, a chronic condition affecting millions worldwide, necessitates early intervention to prevent severe complications. While accurately predicting diabetes onset or progression remains challenging due to complex and imbalanced datasets, recent advancements in machine learning offer potential solutions. Traditional prediction models, often limited by default parameters, have been superseded by more sophisticated approaches. Leveraging Bayesian optimization to fine-tune XGBoost, researchers can harness the power of complex data analysis to improve predictive accuracy. By identifying key factors influencing diabetes risk, personalized prevention strategies can be developed, ultimately enhancing patient outcomes. Successful implementation requires meticulous data management, stringent ethical considerations, and seamless integration into healthcare systems. This study focused on optimizing the hyperparameters of an XGBoost ensemble machine learning model using Bayesian optimization. Compared to grid search XGBoost (accuracy: 97.24%, F1-score: 95.72%, MCC: 81.02%), the XGBoost with Bayesian optimization achieved slightly improved performance (accuracy: 97.26%, F1-score: 95.72%, MCC:81.18%). Although the improvements observed in this study are modest, the optimized XGBoost model with Bayesian optimization represents a promising step towards revolutionizing diabetes prevention and treatment. This approach holds significant potential to improve outcomes for individuals at risk of developing diabetes.
With IoT networks expected to exceed 29 billion connected devices by 2030, the risk of cyberattacks has never been higher. As more devices come online, the attack surface for hackers continues to expand, making cybersecurity a pressing concern. Intrusion Detection Systems (IDS) are essential for identifying and mitigating these threats in real-time. However, a significant challenge IDS faces is dealing with imbalanced datasets, where attack instances are significantly underrepresented compared to normal traffic. Training models on such skewed data leads to a bias toward majority-class patterns, reducing their ability to detect intrusions effectively. To address this issue, this work introduces CSMCR (Cosine Similarity-based Majority Class Reduction), a novel technique that selectively removes redundant majority-class samples while preserving dataset integrity. Unlike traditional approaches like SMOTE (oversampling) or random undersampling, CSMCR ensures that the retained majority instances remain diverse by analyzing feature-wise similarity. This prevents unnecessary data duplication and minimizes information loss. Additionally, we developed a hybrid deep learning model integrating RegNet and FBNet architectures to enhance feature extraction and classification performance. Experimental results on multiple IDS datasets confirm that balancing the dataset to a 1:1 ratio optimally prevents overfitting and improves model interpretability. The proposed model achieved an F1-score of 0.9758 on RT-IoT2022 and 0.9275 on UNSW Bot-IoT, outperforming SMOTE-based methods in accuracy and computational efficiency. Notably, CSMCR reduced training time by 53% compared to conventional oversampling techniques. Incremental training evaluations reveal that bias formation reduces performance beyond a 1:2 majority-to-minority ratio. These findings establish CSMCR as a robust, scalable, and computationally efficient IDS balancing strategy tailored for IoT network security.
International Journal of Computational Intelligence Systems · مجلد 18 (1)
Artificial intelligence (AI) transforms multiple businesses, including medical research, where AI-driven developments bring significant advantages. The application of machine learning algorithms enables medical researchers to examine large amounts of data accurately, which leads to the development of precise and effective treatment approaches. Lung cancer leads the list of critical healthcare issues because it remains the world’s most lethal form of cancer thus demanding innovative diagnostic tools for faster and accurate identification. The proposed study introduces an innovative method called CTGAN-RF which uses conditional tabular generative adversarial networks (CTGAN) and random forest (RF) classifier to detect lung cancer through synthetic data generation. The proposed model demonstrated superior performance by achieving a 0.9893 score of accuracy and 0.99 value for precision, F1 score, and recall. Extensive experimental evaluation for this method included testing nine classification algorithms. The implementation of different classifiers employed data balancing methods including SMOTE and borderline-SMOTE along with SMOTE ENN and unbalanced data configurations. Comparative analysis showed that CTGAN-RF consistently performed significantly better than traditional classifiers in dealing with class imbalance and improving prediction accuracy. After testing with fivefold cross-validation, the reliability of the model was further validated. In comparison to cutting-edge approaches for lung cancer diagnosis, the proposed methodology outperformed in terms of classification metrics. This in-depth evaluation of synthetic data augmentation with machine learning in lung cancer detection has helped in the development of personalized treatment strategies in the fight against such a life-threatening disease.
lung cancer detectionsynthetic data augmentationCTGANrandom forest
Introduction Urban power load forecasting is essential for smart grid planning but is hindered by data imbalance issues. Traditional single-model approaches fail to address this effectively, while multi-model methods mitigate imbalance by splitting datasets but incur high costs and risk losing shared power distribution characteristics. Methods A lightweight urban power load forecasting model (DLUPLF) is proposed, enhancing LSTM networks with contrastive loss in short-term sampling, a difference compensation mechanism, and a shared feature extraction layer to reduce costs. The model adjusts predictions by learning distribution differences and employs dynamic class-center contrastive learning loss for regularization. Its performance was evaluated through parameter tuning and comparative analysis. Results The DLUPLF model demonstrated improved accuracy in forecasting imbalanced datasets while reducing computational costs. It preserved shared power distribution characteristics and outperformed traditional and multi-model approaches in efficiency and prediction accuracy. Discussion DLUPLF effectively addresses data imbalance and model complexity challenges, making it a promising solution for urban power load forecasting. Future work will focus on real-time applications and broader smart grid systems.
IEEE International Conference on Image Processing, Applications and Systems (IPAS) · 1-8
Skin Cancer is the most serious form of cancer, often spreading to other parts of the body over time. Early and accurate identification of skin lesions can improve the treatment options and increase the probability of curing the disease before it progresses. However, automatic detection of skin cancer faces difficulties because of the imbalanced datasets, affected areas covered by hair or moles, and other visual obstructions. In recent years, Deep Learning (DL) techniques have been widely used to address these challenges. Despite advancements, achieving consistently high accuracy remains difficult. To tackle these challenges, the research proposes a novel hybrid DL model for efficient skin cancer detection from dermoscopic images. The proposed network consists of three key modules: feature extraction using the Xception network, feature enhancement through the Squeeze and Excitation network (SENet), and final classification using a transformer module. This combination of DL models effectively identifies SC. The proposed network is evaluated using benign and malignant dermoscopic images collected from the Kaggle website. Necessary preprocessing steps, such as resizing, grayscale conversion, hair removal, and data balancing, are applied. The proposed network is compared with popular DL networks such as MobileNet, Xception, and VGG-16. The experimental outcome demonstrates that the proposed network outperforms these models, achieving a higher accuracy, recall, precision, and F1 score of 96.83%, 97%, 96.67%, and 96.83%, respectively. The metrics value shows that the proposed network is promising and can be helpful for dermatologists in identifying skin cancer with minimal effort.
skin cancer detectiondermoscopic imageshybrid deep learning modelXception
Frontiers in Public Health · مجلد 12 · 1462693-1462693
Fetal health holds paramount importance in prenatal care and obstetrics, as it directly impacts the wellbeing of mother and fetus. Monitoring fetal health through pregnancy is crucial for identifying and addressing potential risks and complications that may arise. Early detection of abnormalities and deviations in fetal health can facilitate timely interventions to mitigate risks and improve outcomes for the mother and fetus. Monitoring fetal health also provides valuable insights into the effectiveness of prenatal interventions and treatments. For fetal health classification, this research work makes use of cardiotocography (CTG) data containing 21 features including fetal growth, development, and physiological parameters such as heart rate and movement patterns with three target classes "normal," "suspect," and "pathological." The proposed methodology makes use of data upsampled using the synthetic minority oversampling technique (SMOTE) to handle the class imbalance problem that is very crucial in medical diagnosing with a light gradient boosting machine. The results show that the proposed model gives 0.9989 accuracy, 0.9988 area under the curve, 0.9832 recall, 0.9834 precision, 0.9832 F1 score, 0.9748 Kappa score, and 0.9749 Matthews correlation coefficient value on the test dataset. The performance of the proposed model is compared with other machine learning models to show the dominance of the proposed model. The proposed model's significance is further evaluated using 10-fold cross-validation and comparing the proposed model with other state-of-the-art models.
fetal health classificationcardiotocographyLightGBMSHAP explainable AI
Background: Differentiated thyroid cancer (DTC) is the most prevalent endocrine malignancy with a recurrence rate of about 20%, necessitating better predictive methods for patient management. This study aims to create a relational classification model to predict DTC recurrence by integrating clinical, pathological, and follow-up data. Methods: The balanced dataset comprises 550 DTC samples collected over 15 years, featuring 13 clinicopathological variables. To address the class imbalance in recurrence status, the Synthetic Minority Over-sampling Technique for Nominal and Continuous (SMOTE-NC) was utilized. A hybrid model combining classification algorithms with association rule mining was developed. Two relational classification approaches, regularized class association rules (RCAR) and classification based on association rules (CBAR), were implemented. Binomial logistic regression analyzed independent predictors of recurrence. Model performance was assessed through accuracy, sensitivity, specificity, positive predictive value, negative predictive value, and F1 score. Results: The RCAR model demonstrated superior performance over the CBAR model, achieving accuracy, sensitivity, and F1 score of 96.7%, 93.1%, and 96.7%, respectively. Association rules highlighted that papillary pathology with an incomplete response strongly predicted recurrence. The combination of incomplete response and lymphadenopathy was also a significant predictor. Conversely, the absence of adenopathy and complete response to treatment were linked to freedom from recurrence. Incomplete structural response was identified as a critical predictor of recurrence risk, even with other low-recurrence conditions. Conclusion: This study introduces a robust and interpretable predictive model that enhances personalized medicine in thyroid cancer care. The model effectively identifies high-risk individuals, allowing for tailored follow-up strategies that could improve patient outcomes and optimize resource allocation in DTC management.
Cervical cancer is one of the deadly diseases that affects women, which requires periodic examinations to identify and treat any cancerous tumors at a preliminary stage. The most prevalent examination tool for cervical cancer prompt identification is the cervical smear (Pap smear) testing; however, due to human negligence, this examination method has an elevated probability of negative findings. Cervical cancer classification using machine learning (ML) and deep learning (DL) has been extensively studied to enhance the conventional diagnostic process. Robust classification results were achieved through the pre-segmented imagery in most current investigations. Conversely, cellular grouping makes reliable cervical cellular segmentation difficult. Additionally, the deep learning methods used in the existing works perform poorly on a multiclass classification when the data distribution is skewed, which is common in the cervical cancer dataset. To mitigate these restrictions in cervical cancer research, this proposed work uses a combination of four different deep-learning methods in various phases of this research. The proposed work is segregated into five phases: pre-processing, data augmentation, segmentation, feature extraction, and classification. Contrast maximization is performed in the pre-processing phase, and the images are augmented using Multi-modal Generative Adversarial Networks (m-GAN) in the second phase. In the third phase, cervical cancer images are segmented using the Seg-UNet model, which is forwarded to the feature extraction phase that employs denoising autoencoders. Finally, the classification is implemented using the Dense CapsNet model and applied to the SIPaKMeD dataset to categorize between normal, abnormal, and benign classes. The proposed system achieves an accuracy of 99.65%, which is higher than the other works in the literature.
cervical cancer classificationPap smeardenoising autoencoderdata augmentation
Addressing the pressing challenge of insurance fraud, which significantly impacts financial losses and trust within the insurance industry, this study introduces an innovative automated detection system utilizing ensemble machine learning (EML) algorithms. The approach encompasses four strategic phases: 1) Tackling data imbalance through diverse re-sampling methods (Over-sampling, Under-sampling, and Hybrid); 2) Optimizing feature selection (Filtering, Wrapping, and Embedding) to enhance model accuracy; 3) employing binary classification techniques (Bagging and Boosting) for effective fraud identification; and 4) applying explanatory model analysis (Shapley Additive Explanations, Break-down plot, and variable-importance Measure) to evaluate the influence of individual features on model performance. Our comprehensive analysis reveals that while not every re-sampling technique improves model performance, all feature selection methods markedly bolster predictive accuracy. Notably, the combination of the Gradient Boosting Machine (GBM) algorithm with NCR re-sampling and GBMVI feature selection emerges as the most effective configuration, offering superior fraud detection capabilities. This study not only advances the theoretical framework for combating insurance fraud through AI but also provides a practical blueprint for insurance companies aiming to incorporate advanced AI strategies into their fraud detection arsenals, thereby mitigating financial risks and fostering trust systems.
IEEE Systems Man and Cybernetics Magazine · مجلد 10 (2) · 32-39
A downsampling strategy based on negative selection density clustering (NSDC-DS) is proposed to improve classifier performance while employing random downsampling for unbalanced communication text. The discovery of self-anomalies via negative selection enhances traditional clustering. The detector and self-set are the sample center point and the sample to be clustered, respectively; anomalous matching is performed on the two; and the NSDC technique analyzes sample similarity. To improve on the traditional downsampling method, we use the Naïve Bayes Support Vector Machine (NBSVM) classifier to identify garbage in sampled communication samples, use principal component analysis (PCA) to evaluate sample information content, propose an improved PCA-signed directed graph (SGD) algorithm to optimize model parameters, and complete semisupervised communication spam text recognition over the Social Web of Things. Several datasets, including unbalanced communication text, were used to compare the improved approach against NSDC, NSDC-DS, PCA-SGD, and standard models. According to the trials, the improved model has a quicker and more consistent convergence speed.
social mediaclass imbalanceprincipal component analysissigned directed graphs
Deep learning and artificial intelligence offer promising tools for improving the accuracy and efficiency of diagnosing various lung conditions using portable chest x-rays (CXRs). This study explores this potential by leveraging a large dataset containing over 6,000 CXR images from publicly available sources. These images encompass COVID-19 cases, normal cases, and patients with viral or bacterial pneumonia. The research proposes a novel approach called "Enhancing COVID Prediction with ESN-MDFS" that utilizes a combination of an Extreme Smart Network (ESN) and a Mean Dropout Feature Selection Technique (MDFS). This study aimed to enhance multi-class lung condition detection in portable chest X-rays by combining static texture features with dynamic deep learning features extracted from a pre-trained VGG-16 model. To optimize performance, preprocessing, data imbalance, and hyperparameter tuning were meticulously addressed. The proposed ESN-MDFS model achieved a peak accuracy of 96.18% with an AUC of 1.00 in a six-fold cross-validation. Our findings demonstrate the model's superior ability to differentiate between COVID-19, bacterial pneumonia, viral pneumonia, and normal conditions, promising significant advancements in diagnostic accuracy and efficiency.
Hepatitis C Virus (HCV) is a viral infection that causes liver inflammation. Annually, approximately 3.4 million cases of HCV are reported worldwide. A diagnosis of HCV in earlier stages helps to save lives. In the HCV review, the authors used a single ML-based prediction model in the current research, which encounters several issues, i.e., poor accuracy, data imbalance, and overfitting. This research proposed a Hybrid Predictive Model (HPM) based on an improved random forest and support vector machine to overcome existing research limitations. The proposed model improves a random forest method by adding a bootstrapping approach. The existing RF method is enhanced by adding a bootstrapping process, which helps eliminate the tree's minor features iteratively to build a strong forest. It improves the performance of the HPM model. The proposed HPM model utilizes a 'Ranker method' to rank the dataset features and applies an IRF with SVM, selecting higher-ranked feature elements to build the prediction model. This research uses the online HCV dataset from UCI to measure the proposed model's performance. The dataset is highly imbalanced; to deal with this issue, we utilized the synthetic minority over-sampling technique (SMOTE). This research performs two experiments. The first experiment is based on data splitting methods, K-fold cross-validation, and training: testing-based splitting. The proposed method achieved an accuracy of 95.89% for k = 5 and 96.29% for k = 10; for the training and testing-based split, the proposed method achieved 91.24% for 80:20 and 92.39% for 70:30, which is the best compared to the existing SVM, MARS, RF, DT, and BGLM methods. In experiment 2, the analysis is performed using feature selection (with SMOTE and without SMOTE). The proposed method achieves an accuracy of 41.541% without SMOTE and 96.82% with SMOTE-based feature selection, which is better than existing ML methods. The experimental results prove the importance of feature selection to achieve higher accuracy in HCV research.
hepatitis C virushybrid predictive modelimproved random forestbootstrapping
International Conference on Inventive Computation Technologies (ICICT) · 112-117
Support vector machine has emerged as one of the most crucial classification techniques in recent years in pattern recognition, machine learning, and data mining. The model training will become more challenging when dealing with multi-classification problems, and its training time will increase considerably as the number of samples increases. A quick training data reduction technique MOIS that works with multiple classification problems is provided to address the issues above. This method takes the cluster centre as the reference point. While removing redundant training samples, selects the decisive boundary samples to significantly reduce the training data and reduce the distribution imbalance between categories. Experimental results show that MOIS can maintain or even improve the classification effect of support vector machines. Moreover, it can significantly improve training efficiency. For example, on the Opt digit data set, the proposed method improves the classification accuracy from 98.94% to 99.05% and, at the same time, shortens the training time to 15%; On the data set composed of the first 100 categories, when the accuracy rate is slightly improved (from 99.29% to 99.30%), the training time is significantly shortened to less than 6% of the original. In addition, MOIS itself has High operating efficiency.
support vector machinemulti-class classificationinstance selectiontraining data reduction