Number Of Input Variables Research Articles

Connectivity conservation planning through deep reinforcement learning

Abstract The United Nations has declared 2021–2030 the decade on ecosystem restoration with the aim of preventing, stopping and reversing the degradation of the ecosystems of the world, often caused by the fragmentation of natural landscapes. Human activities separate and surround habitats, making them too small to sustain viable animal populations or too far apart to enable foraging and gene flow. Despite the need for strategies to solve fragmentation, it remains unclear how to efficiently reconnect nature. In this paper, we illustrate the potential of deep reinforcement learning (DRL) to tackle the spatial optimisation aspect of connectivity conservation planning. The propensity of spatial optimisation problems to explode in complexity depending on the number of input variables and their states is and will continue to be one of its most serious obstacles. DRL is an emerging class of methods focused on training deep neural networks to solve decision‐making tasks and has been used to learn good heuristics for complex optimisation problems. While the potential of DRL to optimise conservation decisions seems huge, only few examples of its application exist. We applied DRL to two real‐world raster datasets in a connectivity planning setting, targeting graph‐based connectivity indices for optimisation. We show that DRL converges to the known optimums in a small example where the objective is the overall improvement of the Integral Index of Connectivity and the only constraint is the budget. We also show that DRL approximates high‐quality solutions on a large example with additional cost and spatial configuration constraints where the more complex Probability of Connectivity Index is targeted. To the best of our knowledge, there is no software that can target this index for optimisation on raster data of this size. DRL can be used to approximate good solutions in complex spatial optimisation problems even when the conservation feature is non‐linear like graph‐based indices. Furthermore, our methodology decouples the optimisation process and the index calculation, so it can potentially target any other conservation feature implemented in current or future software.

Vision-oriented machine learning-assisted seismic energy dissipation estimation for damaged RC beam-column connections

After a seismic event, damage quantification is necessary for safety evaluation, collapse proximity judgment, and measurement of the residual seismic capacity of the joints. Machine learning-based predictive models are developed in this paper aimed at the non-destructive and non-contact estimation of energy dissipation ratio for seismically damaged beam-column joints through surface crack texture complexity indices. For this purpose, the authors collected an experimental dataset comprising 934 images of cracked and/or crushed reinforced concrete-framed joints from 254 specimens under cyclic loading. The associated energy dissipation ratios are then extracted from the lateral load-displacement hysteresis curves for each image by the authors, along with the image-derived fractal indices. Machine learning models with six algorithms, including Linear Regression, Multi-Layer Perceptron, Decision Tree, Random Forest, Gradient Boost, and CatBoost, are trained by 80% of images as a training dataset and validated through the remaining 20% of images in two scenarios. The first scenario develops models by component geometric dimensions and generalized fractal indices calculated for the beams, columns, and joints of seismically damaged connections, while the second scenario adds structural parameters to visual indices to estimate the energy dissipation ratio as the target parameter. Performance evaluation of trained models in the mentioned scenarios via common correlation and accuracy measures, along with the k-fold cross-validation method, indicated that CatBoost-based ML models in both scenarios provide optimal results with R2 = 0.89, RMSE = 0.10 for scenario I and R2 = 0.92, RMSE = 0.09 for scenario II using testing dataset. The first scenario based on the CatBoost algorithm with the R2 variation range of 0.07 and RMSE variation range of 0.03 across the test folds provides a sufficient performance in terms of generalizability and accuracy for estimating the energy dissipation ratio through lower number of input variables.

Augmented Data-Driven Machine Learning for Digital Twin of Stud Shear Connections

Existing design codes for predicting the strength of stud shear connections in composite structures are limited when adapting to constant changes in materials and configurations. Machine learning (ML) models for predicting shear connection are often constrained by the number of input variables, resembling conventional design equations. Moreover, these models tend to overlook considerations beyond those directly comprising the connection. In addition, the data used in ML are often biased and limited in quantity. This study proposes a model using AutoML to automate and optimize the process for predicting the ultimate strength and deformation capacity of shear connections. The proposed model leverages a comprehensive dataset derived from experimental studies and finite element analyses, offering an advanced data-driven solution to overcome the limitations of traditional empirical equations. A digital twin model for the static design of pushout specimens was defined to replace existing empirical design codes. The digital twin model incorporates predictions of the geometry model, ultimate strength, and slip as input parameters and provides criteria for evaluating the limit state through a bilinear load–slip curve. This study advances predictive methodologies in structural engineering by emphasizing the importance of ML in addressing the dynamic and multifaceted nature of shear connection behaviors.

Effect of preconditioning and die thickness on pelleting of livestock feed

ABSTRACT Pelleting has been identified as a means of improving animal performance by limiting selective feeding of livestock. However, it is a complex process that has a number of input variables that must be routinely monitored and adjusted. The interaction between the operational parameters may also give more knowledge of the process. In this study, a screw-type livestock feed pelleting machine was designed and fabricated having a maximum throughput of 0.43 kg/min. Using a full factorial design with three levels of die thickness (5, 7 and 9 mm) and two preconditioners (water and cassava starch), average density of pellets produced was greater than 1g/cm3 For the determination of the density and pellet length, and consequently the durability, the die thickness, plasticiser and interaction of die thickness with plasticiser were significant (p < 0.05). Using starch as preconditioner and increasing the die thickness improved the durability of pellets.

Machine learning for detecting fake accounts and genetic algorithm-based feature selection

Abstract People rely extensively on online social networks (OSNs) in Africa, which aroused cyber attackers’ attention for various nefarious actions. This global trend has not spared African online communities, where the proliferation of OSNs has provided new opportunities and challenges. In Africa, as in many other regions, a burgeoning black-market industry has emerged, specializing in the creation and sale of fake accounts to serve various purposes, both malicious and deceptive. This paper aims to build a set of machine-learning models through feature selection algorithms to predict the fake account, increase performance, and reduce costs. The suggested approach is based on input data made up of features that describe the profiles being investigated. Our findings offer a thorough comparison of various algorithms. Furthermore, compared to machine learning without feature selection and Boruta, machine learning employing the suggested genetic algorithm-based feature selection offers a clear runtime advantage. The final prediction model achieves AUC values between 90% and 99.6%. The findings showed that the model based on the features chosen by the GA algorithm provides a reasonable prediction quality with a small number of input variables, less than 31% of the entire feature space, and therefore permits the accurate separation of fake from real users. Our results demonstrate exceptional predictive accuracy with a significant reduction in input variables using the genetic algorithm, reaffirming the effectiveness of our approach.

A METHOD FOR EVALUATING EXPLANATIONS IN AN ARTIFICIAL INTELLIGENCE SYSTEM USING POSSIBILITY THEORY

The subject of the research is the process of generating explanations for the decision of an artificial intelligence system. Explanations are used to help the user understand the process of reaching the result and to be able to use an intelligent information system more effectively to make practical decisions for him or her. The purpose of this paper is to develop a method for evaluating explanations taking into account differences in input data and the corresponding decision of an artificial intelligence system. The solution of this problem makes it possible to evaluate the relevance of the explanation for the internal decision-making mechanism in an intelligent information system, regardless of the user's level of knowledge about the peculiarities of making and using such a decision. To achieve this goal, the following tasks are solved: structuring the evaluation of explanations depending on their level of detail, taking into account their compliance with the decision-making process in an intelligent system and the level of perception of the user of such a system; developing a method for evaluating explanations based on their compliance with the decision-making process in an intelligent system. Conclusions. The article structures the evaluation of explanations according to their level of detail. The levels of associative dependencies, precedents, causal dependencies and interactive dependencies are identified, which determine different levels of detail of explanations. It is shown that the associative and causal levels of detail of explanations can be assessed using numerical, probabilistic, or possibilistic indicators. The precedent and interactive levels require a subjective assessment based on a survey of users of the artificial intelligence system. The article develops a method for the possible assessment of the relevance of explanations for the decision-making process in an intelligent system, taking into account the dependencies between the input data and the decision of the intelligent system. The method includes the stages of assessing the sensitivity, correctness and complexity of the explanation based on a comparison of the values and quantity of the input data used in the explanation. The method makes it possible to comprehensively evaluate the explanation in terms of resistance to insignificant changes in the input data, relevance of the explanation to the result obtained, and complexity of the explanation calculation. In terms of practical application, the method makes it possible to minimize the number of input variables for the explanation while satisfying the sensitivity constraint of the explanation, which creates conditions for more efficient formation of the interpretation based on the use of a subset of key input variables that have a significant impact on the decision obtained by the intelligent system.

Performance and explainability of feature selection-boosted tree-based classifiers for COVID-19 detection

In this paper, we evaluate the performance and analyze the explainability of machine learning models boosted by feature selection in predicting COVID-19-positive cases from self-reported information. In essence, this work describes a methodology to identify COVID-19 infections that considers the large amount of information collected by the University of Maryland Global COVID-19 Trends and Impact Survey (UMD-CTIS). More precisely, this methodology performs a feature selection stage based on the recursive feature elimination (RFE) method to reduce the number of input variables without compromising detection accuracy. A tree-based supervised machine learning model is then optimized with the selected features to detect COVID-19-active cases. In contrast to previous approaches that use a limited set of selected symptoms, the proposed approach builds the detection engine considering a broad range of features including self-reported symptoms, local community information, vaccination acceptance, and isolation measures, among others. To implement the methodology, three different supervised classifiers were used: random forests (RF), light gradient boosting (LGB), and extreme gradient boosting (XGB). Based on data collected from the UMD-CTIS, we evaluated the detection performance of the methodology for four countries (Brazil, Canada, Japan, and South Africa) and two periods (2020 and 2021). The proposed approach was assessed in terms of various quality metrics: F1-score, sensitivity, specificity, precision, receiver operating characteristic (ROC), and area under the ROC curve (AUC). This work also shows the normalized daily incidence curves obtained by the proposed approach for the four countries. Finally, we perform an explainability analysis using Shapley values and feature importance to determine the relevance of each feature and the corresponding contribution for each country and each country/year.

Utilizing nanotechnology and advanced machine learning for early detection of gastric cancer surgery

Nanotechnology has emerged as a promising frontier in revolutionizing the early diagnosis and surgical management of gastric cancers. The primary factors influencing curative efficacy in GIC patients are drug inefficacy and high surgical and pharmacological therapy recurrence rates. Due to its unique optical features, good biocompatibility, surface effects, and small size effects, nanotechnology is a developing and advanced area of study for detecting and treating cancer. Considering the limitations of GIC MRI and endoscopy and the complexity of gastric surgery, the early diagnosis and prompt treatment of gastric illnesses by nanotechnology has been a promising development. Nanoparticles directly target tumor cells, allowing their detection and removal. It also can be engineered to carry specific payloads, such as drugs or contrast agents, and enhance the efficacy and precision of cancer treatment. In this research, the boosting technique of machine learning was utilized to capture nonlinear interactions between a large number of input variables and outputs by using XGBoost and RNN-CNN as a classification method. The research sample included 350 patients, comprising 200 males and 150 females. The patients' mean ± SD was 50.34 ± 13.04 with a mean age of 50.34 ± 13.04. High-risk behaviors (P = 0.070), age at diagnosis (P = 0.034), distant metastasis (P = 0.004), and tumor stage (P = 0.014) were shown to have a statistically significant link with GC patient survival. AUC was 93.54%, Accuracy 93.54%, F1-score 93.57%, Precision 93.65%, and Recall 93.87% when analyzing stomach pictures. Integrating nanotechnology with advanced machine learning techniques holds promise for improving the diagnosis and treatment of gastric cancer, providing new avenues for precision medicine and better patient outcomes.

Enhancing Airfoil Performance through Artificial Neural Networks and Genetic Algorithm Optimization

As airfoil design plays a crucial role in achieving superior aerodynamic performances, optimization has become an essential part in various engineering applications, including aeronautics and wind energy production. Airfoil optimization using high-fidelity CFD, although highly effective, has proven itself to be time-consuming and computationally expensive. This paper proposes an alternative approach to airfoil performance assessment, through the integration of a deep learning algorithm and a stochastic optimization method. NACA 4-digit parametrization was used for airfoil geometry generation, to ensure feasibility and to reduce the number of input variables. An extensive dataset of airfoil performance parameters has been obtained using an automated CFD solver, laying the foundation for the training of an accurate and robust Artificial Neural Network, capable of accurately predicting aerodynamic coefficients and significantly reducing computational time. Due to the ANN’s predictive capabilities of efficiently navigating vast search spaces, it has been employed as the fitness evaluation method of a multi-objective Genetic Algorithm. Following the optimization process, the resulting airfoils demonstrate significant enhancements in aerodynamic performance and notable improvements in stall behavior. To validate their increased capabilities, a high-fidelity Computational Fluid Dynamics (CFD) validation was conducted. Simulation results demonstrate the approach’s efficacy in finding the optimum airfoil shape for the given conditions and respecting the imposed constraints.

Development and prediction of a robust multivariate trophic state index for the classification of lentic water bodies

This study presents a novel parametric-non parametric composite approach for developing Multivarite Trophic State Index (MTSI) for the classification of lentic water bodies (LWB). The newly developed framework involves the use of water quality parameters (WQPs), Principal Component Analysis (PCA) and Non-Parametric Gaussian Kernal Density (NPGKD) estimator. The generic framework is demonstrated for developing MTSI, based on the datasets of 18 LWBs in Kerala, India. The method could capture the hypereutrophic status and Akkulam lake was found to be in this state and three other lakes are in eutrophic state for the 2012–2018 database used in this study. The method of developing MTSI is generic and with discrete threshold values which enable its application to any hydroclimatic region. Subsequently, a novel Machine Learning based framework is proposed for prediction of MTSI considering WQPs as inputs. Four machine learning methods and the linear regression are applied for the prediction of MTSI of LWBs of Kerala, considering six combinations of input datasets of WQPs. The rigorous performance evaluation based on multitude of error, coefficient and graphical measures confirmed the superiority of Random Forest as the best method for the prediction of MTSI, for the chosen database. The combination of nine WQPs involving both physical and chemical parameters is found to be the best for MTSI prediction and machine learning offers flexibility to get comparable performance even with lesser number of input variables.

The ANN Architecture Analysis: A Case Study on Daylight, Visual, and Outdoor Thermal Metrics of Residential Buildings in China

Selecting an appropriate ANN model is crucial for speeding up the process of building performance simulation during the design phase of residential building layouts, particularly when evaluating three or more green performance metrics simultaneously. In this study, daylight, visual, and outdoor thermal metrics were selected as main green performance. To find the suitable ANN model, sensitivity analysis was used to obtain a set of proper parameters applied to the ANN structure. To train the ANN model with a higher predicting accuracy, this paper tested four different scenarios of ANN parameter setups to find some general guidelines about how to set up an ANN model to predict DF, sunlight hours, QuVue and UTCI. The results showed that an ANN model with a combined output variable demonstrated better average prediction accuracy than ANN models with a separated output variable. Having two times the number of training samplings compared to the number of input variables can lead to a high accuracy of prediction. The ideal number of neurons in the hidden layer was approximately 1.5 times the number of input variables. These findings of how to improve the ANN model may provide guidance for modeling an ANN for building performance.

Data-driven prediction of product yields and control framework of hydrocracking unit

In this study, the relationship between the operating conditions and the product yields and a control framework of the hydrocracking process was developed. The data were collected from a hydrocracking unit in a Chinese refinery. Principal component analysis was used to decrease the number of input variables. Then support vector machine, Gaussian process regression (GPR), and decision tree regression models were developed to establish the relationship above. The best model is GPR, whose Pearson correlation coefficient between the prediction value and the actual value is greater than 0.97 for all the product yields. Shapley additive explanations were performed to interpret the results of the GPR models. A control framework of the hydrocracking unit was then proposed based on the results above. The results show that the machine learning method is a valuable tool for predicting the yield of hydrocracking products, and the control framework proposed helps optimize hydrocracking product yields.

Marriage between variable selection and prediction methods to model plant disease risk

Predicting the risk of a disease in a pathosystem based on a set of climatic variables usually requires handling a high number of input variables, many of which are often irrelevant and/or redundant. Building linear predictive models entails not only dimensionality issues but also the negative impact of multicollinearity. Several feature selection methods have proved to be efficient in both linear and non-linear models, regardless of those issues. However, in a machine learning (ML) context, it is necessary to evaluate these feature selection methods embedded into the model fitting algorithm to obtain the greatest accuracy. The aim of this work was to assess different combinations of variable selection methods with linear and non-linear predictors to fit climate-based models that predict the occurrence of a disease in a pathosystem. Four selection methods were compared: stepwise, which is frequently used in linear models, combined with VIF and p-value statistical criteria (Step+VIF+Pv), and other methods commonly used in ML: filter (F), genetic algorithm (GA), and Boruta (B). The disease risk predictors were constructed with a logistic linear regression model (LR) and the random forest (RF) algorithm, using all the available variables and the subgroups of variables selected by each feature selection method. Data from three pathosystems were processed: two involving Begomovirus –one in common bean (Phaseolus vulgaris L) and the other in soybean (Glycine max)– and the third one involving Mal de Rio Cuarto virus in maize (Zea mays L.). The data sets differed in sample size and number of variables. The accuracy of RF prediction did not vary among feature selection methods. Step+VIF+Pv was used to reduce the model outperformed the other feature selection methods in fitting LR. Our proposal suggests that the appropriate pairing of variable selection and prediction models would improve the modeling of plant disease risk.

Artificial Neural Network for Classification of Dengue Fever Using Backpropagation Algorithm

Fever is an increase in body temperature to higher than usual. Normal human body temperature is at 37oC, if the body temperature is more than this figure, it indicates a fever caused by infectious or non-infectious factors. The main symptom of Dengue hemorrhagic fever is high fever with a temperature between 30oC - 40oC which appears suddenly, the fever lasts for 7 days and occurs continuously, body temperature can be normal or low, then will rise slowly every day and can reach 40oC . These two diseases are still a public health problem in urban areas, including in the cities of Binjai and Medan. The problem that has occurred so far is that people in general cannot differentiate the symptoms of Dengue Fever from Malaria, so the treatment given only provides ordinary fever medicine, so that within three days there is no change and the high body temperature makes the patient know that someone has dengue fever. Therefore, the solution provided in this research is to find out the physical characteristics experienced by the sufferer before further diagnosis is carried out. If someone has a fever above 38oC, the body has red spots, irregular breathing, immediately go to the doctor because these symptoms indicate symptoms of dengue hemorrhagic fever or malaria fever. Artificial neural networks are an information processing system designed to imitate the workings of the human brain by carrying out a learning process through changing the weights of synapses. The human brain consists of millions of interconnected neurons known as biological neurons. Each neuron consists of a cell that has a number of dendrites (input) and an axon (output). Axons connect to other neurons through connecting pathways that produce chemical reactions when responding to incoming input. The input required includes the number of input variables, input variable values, weights, learning rate, threshold, maximum epoh and target (output) with the error value classification used is Mean Absolute Error (MAE), there are 2 types of disease with fever symptoms used. The types of disease are dengue hemorrhagic fever and malaria and the system will be designed using the Visual Basic 2010 programming language. From the results of the research that has been carried out, classification results are obtained with a value of 0.893619481 or rounded to equal 1 and classified as dengue hemorrhagic fever.

Satellite‐Based Fully Connected Neural Network Heating (FCNH) Algorithm for Estimating Latent Heating Rate Inside Storms

AbstractLatent heat (LH) released from precipitation during the water phase change process is the primary energy source driving atmospheric circulation. Current satellite LH retrieval algorithms are mainly physical‐based or lookup table‐based. In this study, a fully connected neural network LH algorithm (FCNH) was developed and tested by weather research and forecasting model (WRF) simulations and global precipitation measurement (GPM) satellite observations. FCNH uses three types of modules: feature representation, feature fusion, and regression. Using satellite observable vertical derivation of precipitation rate () and air temperature (T) as inputs into FCNH achieved the best LH retrieval performance; increasing the number of input variables covering environmental or precipitation characteristics degraded the retrieval accuracy. Compared to the WRF simulated true LH, the FCNH retrieval captured the main features of horizontal and vertical structures with high correlation coefficients and showed improved performance over the associated physical‐based LH algorithm on the same inputs. The FCNH algorithm can alleviate the overestimation of cooling near the surface and the overestimation of positive heating in the mixing layer. The LH retrievals from FCNH and the other three algorithms using inputs of GPM observations were compared, and all achieved basically consistent results. This study is the first attempt to use an artificial neural network method for satellite remote sensing of LH inside precipitation clouds. It promotes understanding of the learning efficiency, accuracy, and limitations of using a fully connected neural network to retrieve LH.

Comparative Analysis of Data-Driven Algorithms for Building Energy Planning via Federated Learning

Building energy planning is a challenging task in the current mounting climate change scenario because the sector accounts for a reasonable percentage of global end-use energy consumption, with a one-fifth share of global carbon emissions. Energy planners rely on physical model-based prediction tools to conserve energy and make decisions towards decreasing energy consumption. For precise forecasting, such a model requires the collection of an enormous number of input variables, which is time-consuming because not all the parameters are easily available. Utilities are reluctant to share retrievable consumer information because of growing concerns regarding data leakage and competitive energy markets. Federated learning (FL) provides an effective solution by providing privacy preserving distributed training to relieve the computational burden and security concerns associated with centralized vanilla learning. Therefore, we aimed to comparatively analyze the effectiveness of several data-driven prediction algorithms for learning patterns from data-efficient buildings to predict the hourly consumption of the building sector in centralized and FL setups. The results provided comparable insights for predicting building energy consumption in a distributed setup and for generalizing to diverse clients. Moreover, such research can benefit energy designers by allowing them to use appropriate algorithms via transfer learning on data of similar features and to learn personalized models in meta-learning approaches.

Application of classification machine learning algorithms for characterizing nutrient transport in a clay plain agricultural watershed

Excess nutrients in surface water and groundwater can lead to water quality deterioration in available water resources. Thus, the classification of nutrient concentrations in water resources has gained significant attention during recent decades. Machine learning (ML) algorithms are considered an efficient tool to describe nutrient loss from agricultural land to surface water and groundwater. Previous studies have applied regression and classification ML algorithms to predict nutrient concentrations in surface water and/or groundwater, or to categorize an output variable using a limited number of input variables. However, there have been no studies that examined the application of different ML classification algorithms in agricultural settings to classify various output variables using a wide range of input variables. In this study, twenty-four ML classification algorithms were implemented on a dataset from three locations within the Upper Parkhill watershed, an agricultural watershed in southern Ontario, Canada. Nutrient concentrations in surface water were classified using geochemical and physical water parameters of surface water and groundwater (e.g., pH), climate and field conditions as the input variables. The performance of these algorithms was evaluated using four evaluation metrics (e.g., classification accuracy) to identify the optimal algorithm for classifying the output variables. Ensemble bagged trees was found to be the optimal ML algorithm for classifying nitrate concentration in surface water (accuracy of 90.9%), while the weighted KNN was the most appropriate algorithm for categorizing the total phosphorus concentration (accuracy of 87%). The ensemble subspace discriminant algorithm gave the highest overall classification accuracy for the concentration of soluble reactive phosphorus and total dissolved phosphorus in surface water with an accuracy of 79.2% and 77.9%, respectively. This study exemplifies that ML algorithms can be used to signify exceedance of recommended concentrations of nutrients in surface waters in agricultural watersheds. Results are useful for decision makers to develop nutrient management strategies.

Modelling of Earphone Design Using Principal Component Analysis

This research investigated a mathematical model of earphone design with principal component analysis. Along with simplifying the design problem, a predictive model for the sound quality indicators, namely, total harmonic distortion, power of output, range of frequency response, signal-to-noise ratio, impedance of the speaker, and headroom, was formulated. (1) Background: Earphone design is a difficult problem requiring excessive experience and know-how in the process. Therefore, this research was developed to formulate a predictive model to facilitate the design process. (2) Methods: A simplified model for the design was developed in previous research, while the sound quality indicators were found to be connected to the eight material-specific parameters. Simultaneously, a principal component analysis (PCA) was utilized to decrease the number of input variables and create a more convenient and streamlined model. (3) Results: The principal component analysis-based approach obtained suboptimal predictive accuracy for the sound quality indicators, but a simplified formulation was obtained. (4) Conclusions: Based on the development and comparison of the modelling approach, it can be seen that principal component analysis can be utilized to simplify the mathematical model of the earphone design problem with a trade-off of accuracy.

Hybridizing machine learning in survival analysis of cardiac PET/CT imaging

BackgroundMachine Learning (ML) allows integration of the numerous variables delivered by cardiac PET/CT, while traditional survival analysis can provide explainable prognostic estimates from a restricted number of input variables. We implemented a hybrid ML-and-survival analysis of multimodal PET/CT data to identify patients who developed myocardial infarction (MI) or death in long-term follow up.MethodsData from 739 intermediate risk patients who underwent coronary CT and selectively stress 15O-water-PET perfusion were analyzed for the occurrence of MI and all-cause mortality. Images were evaluated segmentally for atherosclerosis and absolute myocardial perfusion through 75 variables that were integrated through ML into an ML-CCTA and an ML-PET score. These scores were then modeled along with clinical variables through Cox regression. This hybridized model was compared against an expert interpretation-based and a calcium score-based model.ResultsCompared with expert- and calcium score-based models, the hybridized ML-survival model showed the highest performance (CI .81 vs .71 and .64). The strongest predictor for outcomes was the ML-CCTA score.ConclusionPrognostic modeling of PET/CT data for the long-term occurrence of adverse events may be improved through ML imaging score integration and subsequent traditional survival analysis with clinical variables. This hybridization of methods offers an alternative to traditional survival modeling of conventional expert image scoring and interpretation.

Adapting PINN Models of Physical Entities to Dynamical Data

This article examines the possibilities of adapting approximate solutions of boundary value problems for differential equations using physics-informed neural networks (PINNs) to changes in data about the physical entity being modelled. Two types of models are considered: PINN and parametric PINN (PPINN). The former is constructed for a fixed parameter of the problem, while the latter includes the parameter for the number of input variables. The models are tested on three problems. The first problem involves modelling the bending of a cantilever rod under varying loads. The second task is a non-stationary problem of a thermal explosion in the plane-parallel case. The initial model is constructed based on an ordinary differential equation, while the modelling object satisfies a partial differential equation. The third task is to solve a partial differential equation of mixed type depending on time. In all cases, the initial models are adapted to the corresponding pseudo-measurements generated based on changing equations. A series of experiments are carried out for each problem with different functions of a parameter that reflects the character of changes in the object. A comparative analysis of the quality of the PINN and PPINN models and their resistance to data changes has been conducted for the first time in this study.