Privacy Against Agnostic Inference Attacks in Vertical Federated Learning

Federated Learning (FL) enables decentralized training of machine learning (ML) models, making it a valuable approach for detecting false data in smart power grids (SGs) to enhance grid stability while protecting consumers privacy. However, FL-based ML models remain vulnerable to adversarial attacks during both training and inference phases, which can compromise data security. To address these vulnerabilities, we first investigate the robustness of a novel FL-based false data detection approach using Explainable Artificial Intelligence (XAI), referred to as XAI-based FL detection. This approach utilizes explanations of consumers power consumption data, rather than raw data, during the training process. We assess the robustness of the XAI-based FL detection compared to traditional data-driven FL detection against two types of adversarial attacks: Gradient Inversion attacks in the training phase, where adversaries reconstruct private data from shared gradients, and Evasion attacks in the inference phase, where adversaries subtly modify input data to deceive the detection model. Then, we propose a secure XAI-based FL detector with adversarial training to defend against both attack types. The key idea is that XAI helps mask model gradients during training because XAI-generated explanations remain nearly identical across different samples. Therefore, attackers struggle to accurately reconstruct the original training data, even if they obtain precise explanations using gradient inversion attacks. Additionally, XAI effectively distinguishes between benign and malicious samples. When combined with adversarial training, XAI strengthens model robustness against evasion attacks without compromising accuracy, effectively resolving the trade-off between security and performance. Our proposed detector reduced the success rate of evasion attacks from 94.99% to 29.11 explanations, and further to 0% with adding adversarial training. It also increased the mean square error for gradient inversion attacks from 0.01 to 2.60 in the most severe attack scenarios, making such attacks ineffective.

Induction motors (IMs) are widely used in industrial applications due to their advantages over other motor types. However, the efficiency and lifespan of IMs can be significantly impacted by operating conditions, especially Unbalanced Supply Voltages (USV), which are common in industrial plants. Detecting and accurately assessing the severity of USV in real-time is crucial to prevent major breakdowns and enhance reliability and safety in industrial facilities. This paper presented a reliable method for precise online detection of USV by monitoring a relevant indicator, denominated by negative voltage factor (NVF), which, in turn, is obtained using the voltage symmetrical components. On the other hand, impedance estimation proves to be fundamental to understand the behavior of motors and identify possible problems. IM impedance affects its performance, namely torque, power factor and efficiency. Furthermore, as the presence of faults or abnormalities is manifested by the modification of the IM impedance, its estimation is particularly useful in this context. This paper proposed two machine learning (ML) models, the first one estimated the IM stator phase impedance, and the second one detected USV conditions. Therefore, the first ML model was capable of estimating the IM phases impedances using just the phase currents with no need for extra sensors, as the currents were used to control the IM. The second ML model required both phase currents and voltages to estimate NVF. The proposed approach used a combination of a Regressor Decision Tree (DTR) model with the Short Time Least Squares Prony (STLSP) technique. The STLSP algorithm was used to create the datasets that will be used in the training and testing phase of the DTR model, being crucial in the creation of both features and targets. After the training phase, the STLSP technique was again used on completely new data to obtain the DTR model inputs, from which the ML models can estimate desired physical quantities (phases impedance or NVF).

ObjectiveThis retrospective, case-control study with internal validation evaluates the performance of machine learning (ML) and deep learning (DL) models in classifying pediatric patients at risk for anxiety disorders using structured electronic health records (EHRs) and area-based measures of health (ABMH). The aim is to enable proactive care by monitoring potential anxiety onset across developmental stages.MethodsWe trained a series of ML models (Logistic Regression, Decision Tree, Random Forest, K-Nearest Neighbors, XGBoost) and DL models (LSTM, GRU, RETAIN, Dipole) using structured EHR data from 30-day windows prior to diagnosis. Two datasets were used per age group: one with structured EHR data only, and another including both EHR and ABMH data. ML models were trained using short-term cross-sectional features, while DL models leveraged full longitudinal patient histories. Performance was assessed using AUROC, AUPRC, PPV, NPV, F1 score, and accuracy. Due to differences in input scope, model performance reflects both algorithmic and temporal design differences and is not intended as a direct comparison between ML and DL.ResultsML models offered strong baseline performance, with XGBoost achieving AUROC scores of 0.817 (EHR) and 0.816 (EHR+ABMH) for 8-year-olds. Adding ABMH features did not significantly improve performance. DL models, particularly RETAIN and Dipole, achieved the highest AUROC values (e.g., Dipole: 0.853 with EHR, 0.857 with EHR+ABMH for 8-year-olds), outperforming other DL and ML models within their respective design constraints.ConclusionBoth ML and DL models successfully identified likely anxiety onset using structured EHR data. DL models using longitudinal data achieved the highest performance, while XGBoost provided a robust ML baseline. The minimal impact of ABMH features highlights integration challenges, and performance variation across ages emphasizes the need for age-stratified modeling approaches.

Deep neural networks (DNNs) have shown success in image classification, with high accuracy in recognition of everyday objects. Performance of DNNs has traditionally been measured assuming human accuracy is perfect. In specific problem domains, however, human accuracy is less than perfect and a comparison between humans and machine learning (ML) models can be performed. In recognising everyday objects, humans have the advantage of a lifetime of experience, whereas DNN models are trained only with a limited image dataset. We have tried to compare performance of human learners and two DNN models on an image dataset which is novel to both, i.e. histological images. We thus aim to eliminate the advantage of prior experience that humans have over DNN models in image classification. Ten classes of tissues were randomly selected from the undergraduate first year histology curriculum of a Medical School in North India. Two machine learning (ML) models were developed based on the VGG16 (VML) and Inception V2 (IML) DNNs, using transfer learning, to produce a 10-class classifier. One thousand (1000) images belonging to the ten classes (i.e. 100 images from each class) were split into training (700) and validation (300) sets. After training, the VML and IML model achieved 85.67 and 89% accuracy on the validation set, respectively. The training set was also circulated to medical students (MS) of the college for a week. An online quiz, consisting of a random selection of 100 images from the validation set, was conducted on students (after obtaining informed consent) who volunteered for the study. 66 students participated in the quiz, providing 6557 responses. In addition, we prepared a set of 10 images which belonged to different classes of tissue, not present in training set (i.e. out of training scope or OTS images). A second quiz was conducted on medical students with OTS images, and the ML models were also run on these OTS images. The overall accuracy of MS in the first quiz was 55.14%. The two ML models were also run on the first quiz questionnaire, producing accuracy between 91 and 93%. The ML models scored more than 80% of medical students. Analysis of confusion matrices of both ML models and all medical students showed dissimilar error profiles. However, when comparing the subset of students who achieved similar accuracy as the ML models, the error profile was also similar. Recognition of ‘stomach’ proved difficult for both humans and ML models. In 04 images in the first quiz set, both VML model and medical students produced highly equivocal responses. Within these images, a pattern of bias was uncovered–the tendency of medical students to misclassify ‘liver’ tissue. The ‘stomach’ class proved most difficult for both MS and VML, producing 34.84% of all errors of MS, and 41.17% of all errors of VML model; however, the IML model committed most errors in recognising the ‘skin’ class (27.5% of all errors). Analysis of the convolution layers of the DNN outlined features in the original image which might have led to misclassification by the VML model. In OTS images, however, the medical students produced better overall score than both ML models, i.e. they successfully recognised patterns of similarity between tissues and could generalise their training to a novel dataset. Our findings suggest that within the scope of training, ML models perform better than 80% medical students with a distinct error profile. However, students who have reached accuracy close to the ML models, tend to replicate the error profile as that of the ML models. This suggests a degree of similarity between how machines and humans extract features from an image. If asked to recognise images outside the scope of training, humans perform better at recognising patterns and likeness between tissues. This suggests that ‘training’ is not the same as ‘learning’, and humans can extend their pattern-based learning to different domains outside of the training set.

Simple SummaryEndoscopic resection (ER) is a treatment option for clinically T1a early gastric cancer (EGC) without suspicion of lymph node metastasis (LNM). In patients with non-curative resection after ER, additional surgery is recommended owing to the LNM risk. However, of those patients treated with additional surgery after ER, the actual rate of LNM was about 5–10%; that is, the other patients underwent unnecessary surgeries. Therefore, it is crucial to estimate LNM risk in EGC patients to determine additional management after ER. We derived a machine learning (ML) model to stratify the LNM risk in EGC patients and validate its performance. The constructed ML model, which showed good performance with an area under the receiver operating characteristic of 0.85 or higher, could stratify LNM risk into very low (<1%), low (<3%), intermediate (<7%), and high (≥7%) risk categories. These findings suggest that the ML model can stratify the LNM risk in EGC patients.Stratification of the risk of lymph node metastasis (LNM) in patients with non-curative resection after endoscopic resection (ER) for early gastric cancer (EGC) is crucial in determining additional treatment strategies and preventing unnecessary surgery. Hence, we developed a machine learning (ML) model and validated its performance for the stratification of LNM risk in patients with EGC. We enrolled patients who underwent primary surgery or additional surgery after ER for EGC between May 2005 and March 2021. Additionally, patients who underwent ER alone for EGC between May 2005 and March 2016 and were followed up for at least 5 years were included. The ML model was built based on a development set (70%) using logistic regression, random forest (RF), and support vector machine (SVM) analyses and assessed in a validation set (30%). In the validation set, LNM was found in 337 of 4428 patients (7.6%). Among the total patients, the area under the receiver operating characteristic (AUROC) for predicting LNM risk was 0.86 in the logistic regression, 0.85 in RF, and 0.86 in SVM analyses; in patients with initial ER, AUROC for predicting LNM risk was 0.90 in the logistic regression, 0.88 in RF, and 0.89 in SVM analyses. The ML model could stratify the LNM risk into very low (<1%), low (<3%), intermediate (<7%), and high (≥7%) risk categories, which was comparable with actual LNM rates. We demonstrate that the ML model can be used to identify LNM risk. However, this tool requires further validation in EGC patients with non-curative resection after ER for actual application.

Introduction: Air pollution is a significant global health challenge, contributing to the deaths of millions of people annually. Among these pollutants, Particulate Matter (PM2.5) is the most harmful to the respiratory system causing serious health problems. This study focused on predicting PM2.5 in the air of Islamabad, capital of Pakistan by using machine learning and deep learning models. Materials and methods: Two machine learning models (Decision Tree and Random Forest) and four deep learning models including Multi-Layer Neural Network (MLNN), Long Short-Term Memory (LSTM), Recurrent Neural Network (RNN), Gated Recurrent Unit (GRU) are used in the study. Each model's performance was assessed by using statistical indicators including coefficient of determination (R2), Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Relative Root Mean Square Error (RRMSE). These models are also ranked based on their performance by compromise programming technique. Results: Machine learning models performed better in the training phase by achieving higher R2 values of 0.98 and 0.97 but couldn’t maintain the same performance in the testing phase. Whereas the deep learning models performed best in both the training and testing phases. MLNN model attained higher R2 value of 0.98 in training and 0.88 in testing and is evaluated as top-ranked prediction model in predicting particulate matter PM2.5. Whereas,LSTM, GRU, RNN, Decision Tree, and Random Forest are placed at the 2nd,3rd, 4th, 5th, and 6th positions having R2 values of 0.86, 0.87, 0.82, 0.99, and0.97 during training and 0.71, 0.69, 0.69, 0.75, and 0.85 respectively during testing. Conclusion: Deep learning models, especially MLNN, showed strong performance in predicting PM2.5 as compared to the machine learning models.

IntroductionPercutaneous coronary intervention (PCI) has been the main treatment of coronary artery disease (CAD). In this review, we aimed to compare the performance of machine learning (ML) vs. logistic regression (LR) models in predicting different outcomes after PCI.MethodsStudies using ML or deep learning (DL) models to predict mortality, MACE, in-hospital bleeding, and acute kidney injury (AKI) after PCI or primary PCI were included. Articles were excluded if they did not provide a c-statistic, solely used ML models for feature selection, were not in English, or only used logistic or LASSO regression models. Best-performing ML and LR-based models (LR model or conventional risk score) from the same studies were pooled separately to directly compare the performance of ML versus LR. Risk of bias was assessed using the PROBAST and CHARMS checklists.ResultsA total of 59 studies were included. Meta-analysis showed that ML models resulted in a higher c-statistic compared to LR in long-term mortality (0.84 vs. 0.79, P-value = 0.178), short-term mortality (0.91 vs. 0.85, P = 0.149), bleeding (0.81 vs. 0.77 P = 0.261), acute kidney injury (AKI; 0.81 vs. 0.75, P = 0.373), and major adverse cardiac events (MACE; 0.85 vs. 0.75, P = 0.406). PROBAST analysis showed that 93% of long-term mortality, 70% of short-term mortality, 89% of bleeding, 69% of AKI, and 86% of MACE studies had a high risk of bias.ConclusionNo statistical significance existed between ML and LR model. In addition, the high risk of bias in ML studies and complexity in interpretation undermines their validity and may impact their adaption in a clinical settings.

BackgroundEarly prediction of hospital mortality is crucial for ICU patients with sepsis. This study aimed to develop a novel blending machine learning (ML) model for hospital mortality prediction in ICU patients with sepsis.MethodsTwo ICU databases were employed: eICU Collaborative Research Database (eICU-CRD) and Medical Information Mart for Intensive Care III (MIMIC-III). All adult patients who fulfilled Sepsis-3 criteria were identified. Samples from eICU-CRD constituted training set and samples from MIMIC-III constituted test set. Stepwise logistic regression model was used for predictor selection. Blending ML model which integrated nine sorts of basic ML models was developed for hospital mortality prediction in ICU patients with sepsis. Model performance was evaluated by various measures related to discrimination or calibration.ResultsTwelve thousand five hundred fifty-eight patients from eICU-CRD were included as the training set, and 12,095 patients from MIMIC-III were included as the test set. Both the training set and the test set showed a hospital mortality of 17.9%. Maximum and minimum lactate, maximum and minimum albumin, minimum PaO2/FiO2 and age were important predictors identified by both random forest and extreme gradient boosting algorithm. Blending ML models based on corresponding set of predictors presented better discrimination than SAPS II (AUROC, 0.806 vs. 0.771; AUPRC 0.515 vs. 0.429) and SOFA (AUROC, 0.742 vs. 0.706; AUPRC 0.428 vs. 0.381) on the test set. In addition, calibration curves showed that blending ML models had better calibration than SAPS II.ConclusionsThe blending ML model is capable of integrating different sorts of basic ML models efficiently, and outperforms conventional severity scores in predicting hospital mortality among septic patients in ICU.

For recent years, Machine Learning (ML) models have been proven to be useful in solving problems of a wide variety of fields such as medical, economic, manufacturing, transportation, energy, education, etc. With increased interest in ML models and advances in sensor technologies, ML models are being widely applied even in civil engineering domain. ML model enables analysis of large amounts of data, automation, improved decision making and provides more accurate prediction. While several state-of-the-art reviews have been conducted in each sub-domain (e.g., geotechnical engineering, structural engineering) of civil engineering or its specific application problems (e.g., structural damage detection, water&amp;#160;quality evaluation), little effort has been devoted to comprehensive review on ML models applied in civil engineering and compare them across sub-domains. A systematic, but domain-specific literature review framework should be employed to effectively classify and compare the models. To that end, this study proposes a novel review approach based on the hierarchical classification tree &amp;#8220;D-A-M-I-E (Domain-Application problem-ML models-Input data-Example case)&amp;#8221;. &amp;#8220;D-A-M-I-E&amp;#8221; classification tree classifies the ML studies in civil engineering based on the (1) domain of the civil engineering, (2) application problem, (3) applied ML models and (4) data used in the problem. Moreover, data used for the ML models in each application examples are examined based on the specific characteristic of the domain and the application problem. For comprehensive review, five different domains (structural engineering, geotechnical engineering, water engineering, transportation engineering and energy engineering) are considered and the ML application problem is divided into five different problems (prediction, classification, detection, generation, optimization). Based on the &amp;#8220;D-A-M-I-E&amp;#8221; classification tree, about 300 ML studies in civil engineering are reviewed. For each domain, analysis and comparison on following questions has been conducted: (1) which problems are mainly solved based on ML models, (2) which ML models are mainly applied in each domain and problem, (3) how advanced the ML models are and (4) what kind of data are used and what processing of data is performed for application of ML models. This paper assessed the expansion and applicability of the proposed methodology to other areas (e.g., Earth system modeling, climate science). Furthermore, based on the identification of research gaps of ML models in each domain, this paper provides future direction of ML in civil engineering based on the approaches of dealing data (e.g., collection, handling, storage, and transmission) and hopes to help application of ML models in other fields.

Machine learning is a major area in artificial intelligence, which enables computer to learn itself explicitly without programming. As machine learning is widely used in making decision automatically, attackers have strong intention to manipulate the prediction generated my machine learning model. In this paper we study about the different types of attacks and its countermeasures on machine learning model. By research we found that there are many security threats in various algorithms such as K-nearest-neighbors (KNN) classifier, random forest, AdaBoost, support vector machine (SVM), decision tree, we revisit existing security threads and check what are the possible countermeasures during the training and prediction phase of machine learning model. In machine learning model there are 2 types of attacks that is causative attack which occurs during the training phase and exploratory attack which occurs during the prediction phase, we will also discuss about the countermeasures on machine learning model, the countermeasures are data sanitization, algorithm robustness enhancement, and privacy preserving techniques.

Machine learning(ML) models today are vulnerable to several types of attacks. In this work, we will study a category of attack known as membership inference attack and show how ML models are susceptible to leaking secure information under such attacks. Given a data record and a black box access to a ML model, we present a framework to deduce whether the data record was part of the model’s training dataset or not. We achieve this objective by creating an attack ML model which learns to differentiate the target model’s predictions on its training data from target model’s predictions on data not part of its training data. In other words, we solve this membership inference problem by converting it into a binary classification problem. We also study mitigation strategies to defend the ML models against the attacks discussed in this work. In this paper evaluation method on real world datasets: (1) CIFAR-10 and (2) UCI Adult (Census Income) using classification as the task performed by the target ML models built on these datasets.

Machine learning (ML) and deep learning (DL) models can play an important role when it comes to modelling complicated processes. Such capability is necessary for hydrological and climate-related applications. Generally, ML models utilize precipitation and temperature time series of a basin as input to develop a lumped rainfall-runoff model to simulate streamflow at the basin outlet. However, when it is divided into several sub-basins, Graph Neural Networks (GNN) can consider each sub-basin as a node and link them together using a connectivity matrix to account for spatial variations of hydroclimatic variables. In this study, GNN and various ML models with different types of architecture, ranging from neural networks, tree-based structure, and gradient boosting, were exploited for daily streamflow simulation over different case studies. For each case study, the basin was divided into a few sub-basins for which daily precipitation and temperature data were aggregated and used as input. For training GNN, the connection matrix of sub-basins was also used as input. Basically, 75% of historical records were utilized to train GNN and different ML models, e.g., artificial neural networks, support vector machine, decision tree, random forest, eXtreme Gradient Boosting (XGBoost), Light Gradient-Boosting Machine (LightGBM), and Category Boosting (CatBoost), while the rest was used for testing. Streamflow simulation was conducted with/without considering seasonality impact and lag times. The obtained results clearly demonstrate that considering seasonality and time lags can enhance accuracy of streamflow predictions based on Kling–Gupta efficiency (KGE). Furthermore, GNN with seasonality impact and time lags achieved promising results across different case studies with KGE&gt;0.85 for training and KGE&gt;0.59 for testing data, respectively. Among ML models, boosting models, e.g., LightGBM and XGBoost, performed slightly better than other ML models. for Finally, this comparative analysis provides valuable insights for ML/DL applications in climate change impact assessments.Acknowledgements: This research work was carried out as part of the TRANSCEND project with funding received from the European Union Horizon Europe Research and Innovation Programme under Grant Agreement No. 10108411.

Machine learning-based prediction of abdominal aortic aneurysms for individualized patient care

Necrotizing enterocolitis (NEC) is one of the most common diseases that pose serious threats to the life of newborns. In clinical practice, NEC is typically treated by surgical intervention, but it is still difficult to identify the timing of surgical intervention for this disease. Therefore, this study was conducted to establish a machine learning (ML) model for identifying the optimal timing of surgical intervention for NEC by comparing logistic regression (LR) models with ML models and to visualize important influencing indicators via a nomogram. The basic information, clinical manifestations, laboratory examination results, and radiography imaging results of newborns who were diagnosed with NEC in Qilu Hospital of Shandong University from 2011 to 2024 were collected and processed. Besides, some specific indicators were screened using univariate and multivariate LR analysis and ML analysis methods (including the random forest [RF] algorithm, support vector machine [SVM], decision tree [DT], naive Bayes [NB], and k-nearest Neighbor [KNN]) to construct a clinical model to predict the timing of surgical intervention for NEC. Moreover, a nomogram for predicting the timing of surgical intervention for NEC was constructed based on the independent risk factors selected by the multivariate LR analysis. Finally, the performance of each ML model was evaluated by the receiver operating characteristic (ROC) curve, calibration curve, and decision curve analysis (DCA). A total of four differential indicators related to surgical intervention for NEC were screened by univariate and multivariate LR analyses. The five ML models were evaluated according to these indicators and then compared with a classical LR model. The results demonstrated that the LR model exhibited the best performance. Among the five ML models, the RF model displayed the best overall performance. In addition, a nomogram was plotted according to the LR analysis results to visualize the scores of important indicators. The results revealed that interloop space widening had the highest score. The indicator evaluation results and the analysis results based on ROC curves, DCA curves, and calibration curves corroborate that the LR model as a classical model achieves the best performance. In addition to the LR model, the RF model displays excellent performance among the five ML models. Therefore, it is expected to use this ML model to identify a more suitable surgical timing for newborns with NEC.

To develop and identify machine learning (ML) models using pretreatment clinical and 2-deoxy-2-[18F]fluoro-D-glucose positron emission tomography ([18F]-FDG-PET)-based radiomic characteristics to predict disease recurrences in patients with breast cancers who underwent surgery. This retrospective study included 112 patients with 118 breast cancer lesions who underwent [18F]-FDG-PET/ X-ray computed tomography (CT) preoperatively, and these lesions were assigned to training (n=95) and testing (n=23) cohorts. A total of 12 clinical and 40 [18F]-FDG-PET-based radiomic characteristics were used to predict recurrences using 7 different ML algorithms, namely, decision tree, random forest (RF), neural network, k-nearest neighbors, naive Bayes, logistic regression, and support vector machine (SVM) with a 10-fold cross-validation and synthetic minority over-sampling technique. Three different ML models were created using clinical characteristics (clinical ML models), radiomic characteristics (radiomic ML models), and both clinical and radiomic characteristics (combined ML models). Each ML model was constructed using the top ten characteristics ranked by the decrease in Gini impurity. The areas under ROC curves (AUCs) and accuracies were used to compare predictive performances. In training cohorts, all 7 ML algorithms except for logistic regression algorithm in the radiomics ML model (AUC = 0.760) achieved AUC values of >0.80 for predicting recurrences with clinical (range, 0.892-0.999), radiomic (range, 0.809-0.984), and combined (range, 0.897-0.999) ML models. In testing cohorts, the RF algorithm of combined ML model achieved the highest AUC and accuracy (95.7% (22/23)) with similar classification performance between training and testing cohorts (AUC: training cohort, 0.999; testing cohort, 0.992). The important characteristics for modeling process of this RF algorithm were radiomic GLZLM_ZLNU and AJCC stage. ML analyses using both clinical and [18F]-FDG-PET-based radiomic characteristics may be useful for predicting recurrence in patients with breast cancers who underwent surgery.