High Prediction Efficiency Research Articles

Deep Learning Neural Network Model for Tunnel Ground Surface Settlement Prediction Based on Sensor Data

Monitoring and prediction of ground settlement during tunnel construction are of great significance to ensure the safe and reliable operation of urban tunnel systems. Data-driven techniques combining artificial intelligence (AI) and sensor networks are popular methods in the field, which have several advantages, including high prediction accuracy, efficiency, and low cost. Deep learning, as one of the advanced techniques in AI, is demanded for the tunnel settlement forecasting problem. However, deep neural networks often require a large amount of training data. Due to the tunnel construction, the available training data samples are limited, and the data are univariate (i.e., containing only the settlement data). In response to the above problems, this research proposes a deep learning model that only requires limited number of training data for short-period prediction of the tunnel surface settlement. In the proposed complete ensemble empirical mode decomposition with adaptive noise long short term memory (CEEMDAN-LSTM model), single-dimensional data is divided into multidimensional data by CEEMDAN through the complete ensemble empirical mode decomposition. Each component is then predicted by a LSTM neural network and superimposed for obtaining the final prediction result. Experimental results show that, compared with existing machine learning techniques and algorithms, this deep learning method has higher prediction accuracy and acceptable computational efficiency. In the case of small samples, this method can significantly improve the accuracy of time series forecasting.

A novel compliant assembly variation analysis with consideration of initial deviation and welding shrinkage

Welding, as the most common joining technique, causes welding deformations that reduce the geometrical accuracy of hull structures. In conventional variation analysis, welding distortion is not taken into account. In this paper, a compliant assembly variation modeling and analysis method is presented to investigate the impact of local weld-induced shrinkage on the propagation relation from the initial deviation to the final assembly deviation in thin plate assembly. First, welding deformations in the form of inherent strains are incorporated into variation analysis model based on Kirchhoff plate model. Additionally, part deviations are also introduced via equivalent forces in order to consider the effects of geometrical variation on compliant assembly. Then, other variation sources, such as fixture locating errors, are modeled as equality constraints of the mechanical variation model, which are then transformed into boundary conditions with Lagrange multiplier method. With an additional stiffness matrix, we describe concisely the coupling relationship between initial deviations and welding deformations caused by gaps and misalignments. In order to validate the effectiveness of the proposed method, numerical simulations and experimental verifications are carried out for two assembly cases, and the results show that the present method has high prediction accuracy and efficiency. This study provides a potential application for improving assembly dimensions considering the inevitable manufacturing deviations and welding deformations.

Landslide susceptibility mapping with the fusion of multi-feature SVM model based FCM sampling strategy: A case study from Shaanxi Province

ABSTRACT The quality of “non-landslide’ samples data impacts the accuracy of geological hazard risk assessment. This research proposed a method to improve the performance of support vector machine (SVM) by perfecting the quality of ‘non-landslide’ samples in the landslide susceptibility evaluation model through fuzzy c-means (FCM) cluster to generate more reliable susceptibility maps. Firstly, three sample selection scenarios for ‘non-landslide’ samples include the following principles: 1) select randomly from low-slope areas (scenario-SS), 2) select randomly from areas with no hazards (scenario-RS), 3) obtain samples from the optimal FCM model (scenario-FCM), and then three sample scenarios are constructed with 10,193 landslide positive samples. Next, we have compared and evaluated the performance of three sample scenarios in the SVM models based on the statistical indicators such as the proportion of disaster points, density of disaster points precision, receiver operating characteristic (ROC) curve, and area under the ROC curve (AUC). Finally, The evaluation results show that the ‘non-landslide’ negative samples based on the FCM model are more reasonable. Furthermore, the hybrid method supported by SVM and FCM models exhibits the highest prediction efficiency. Scenario FCM produces an overall accuracy of approximately 89.7% (AUC), followed by scenario-SS (86.7%) and scenario-RS (85.6%).

Experimental and Theoretical Study of the Axial Distribution of Solid Phase Particles in a Fluidized Bed

The article presents the results of computational and experimental studies of the distribution of a model material (plastic spherical particles with a size of 6 mm) along the height of a laboratory two-dimensional apparatus of the fluidized bed of the periodic principle of action. To experimentally determine the distribution of the solid phase over the height of the apparatus, digital photographs of the fluidized bed were taken, which were then analyzed using an algorithm that had been specially developed for this purpose. The algorithm involved splitting the image by height into separate rectangular areas, identifying the particles and counting their number in each of these areas. Numerical experiments were performed using the previously proposed one-dimensional cell model of the fluidization process, constructed on the basis of the mathematical apparatus of the theory of Markov chains with discrete space and time. The design scheme of the model assumes the spatial decomposition of the layer in height into individual elements of small finite sizes. Thus, the numerically obtained results qualitatively corresponded to the full-scale field experiment that had been set up. To ensure the quantitative reliability of the calculated forecasts, a parametric identification of the model was performed using known empirical dependencies to calculate the particle resistance coefficient and estimate the coefficient of their macrodiffusion. A comparison of the results of numerical and field experiments made us possible to identify the most productive empirical dependencies that correspond to the cellular scheme of modeling the process. The resulting physical and mathematical model has a high predictive efficiency and can be used for engineering calculations of devices with a fluidized bed, as well as for setting and solving problems of optimal control of technological processes in these devices for various target functions.

Predictive analytics as a tool of controlling in decision making process in the marina industry

This paper is dealing with predictive modeling based on predictive analytics using computer application system and the usage of the prediction results for decision-making processes. Usually the prediction is based on the experience of decision makers, but the aim of this study is to explain and proof higher predictive efficiency when using predictive analytics based on machine learning as well as more accurate future-oriented business decisions. The marina industry in Croatia is used for this research because of its complexity and necessity to predict future events that influence company success with reliable accuracy. The information for decision-making were obtained from the customer database recorded manually over the past 30 years and according to data from December 2020. The optimized prediction by the vector machine and statistical theory based on the Bayes theorem is used to support more accurate prediction. The quantitative research was carried out using the SAP Predictive Analytics (SAP PA) computer application. The results of prediction models are a perfect basis for making future-oriented strategic and tactical decisions. This research proves that, with knowledge obtained from the results of prediction models it is possible to improve the identification of the target group among applicants and customers that contribute to company success. The research provides a theoretical and an empirical contribution in the usage of predictive analytics in the marina industry in Croatia.

Management forecast based on big data fusion DEA and RBF algorithm

with the rapid development of information technology, how to use big data for effective analysis has become the focus of various industries. Big data will inevitably bring data noise and data redundancy. The traditional big data preprocessing methods do not take into account the functional relationship between variables, and the commonly used big data modeling tools can not effectively model big data with complex nonlinear relationships. This paper puts forward a prediction method combining DEA and RBF, uses DEA to preprocess big data to select the effective date, then uses RBF to model the data, and compares the prediction accuracy with the original environmental management fund data without pre-processing. The results show that the DEA-RBF method has higher prediction efficiency and better prediction accuracy.

Modeling the Dynamics of Radiation Belt Electrons With Source and Loss Driven by the Solar Wind

AbstractA radial diffusion model directly driven by the solar wind is developed to reproduce MeV electron variations between L = 2–12 (L is L* in this study) from October 2012 to April 2015. The radial diffusion coefficient, internal source rate, quick loss due to EMIC waves, and slow loss due to hiss waves are all expressed in terms of the solar wind speed, dynamic pressure, and interplanetary magnetic field (IMF). The model achieves a prediction efficiency (PE) of 0.45 at L = 5 and 0.51 at L = 4 after converting the electron phase space densities to differential fluxes and comparing with Van Allen Probes measurements of 2 and 3 MeV electrons at L = 5 and L = 4, respectively. Machine learning techniques are used to tune parameters to get higher PE. By tuning parameters for every 60‐day period, the model obtains PE values of 0.58 and 0.82 at L = 5 and L = 4, respectively. Inspired by these results, we divide the solar wind activity into three categories based on the condition of solar wind speed, IMF Bz, and dynamic pressure, and then tune these three sets of parameters to obtain the highest PE. This experiment confirms that the solar wind speed has the greatest influence on the electron flux variations, particularly at higher L, while the dynamic pressure has more influence at lower L. Also, the PE at L = 4 is mostly higher than those at L = 5, suggesting that the electron loss due to the magnetopause shadowing combined with the outward radial diffusion is not well captured in the model.

Comparison of Different Machine Models Based on Contrast-Enhanced Computed Tomography Radiomic Features to Differentiate High From Low Grade Clear Cell Renal Cell Carcinomas.

PurposeThis study was to investigate the role of different radiomics models with enhanced computed tomography (CT) scan in differentiating low from high grade renal clear cell carcinomas.Materials and MethodsCT data of 190 cases with pathologically confirmed renal cell carcinomas were collected and divided into the training set and testing set according to different time periods, with 122 cases in the training set and 68 cases in the testing set. The region of interest (ROI) was delineated layer by layer.ResultsA total of 402 radiomics features were extracted for analysis. Six of the radiomic parameters were deemed very valuable by univariate analysis, rank sum test, LASSO cross validation and correlation analysis. From these six features, multivariate logistic regression model, support vector machine (SVM), and decision tree model were established for analysis. The performance of each model was evaluated by AUC value on the ROC curve and decision curve analysis (DCA). Among the three prediction models, the SVM model showed a high predictive efficiency. The AUC values of the training set and the testing set were 0.84 and 0.83, respectively, which were significantly higher than those of the decision tree model and the multivariate logistic regression model. The DCA revealed a better predictive performance in the SVM model that possessed the highest degree of coincidence.ConclusionRadiomics analysis using the SVM radiomics model has highly efficiency in discriminating high- and low-grade clear cell renal cell carcinomas.

Computed-Tomography-Based Radiomics Model for Predicting the Malignant Potential of Gastrointestinal Stromal Tumors Preoperatively: A Multi-Classifier and Multicenter Study.

Background: Our goal was to establish and verify a radiomics risk grading model for gastrointestinal stromal tumors (GISTs) and to identify the optimal algorithm for risk stratification.Methods: We conducted a retrospective analysis of 324 patients with GISTs, the presence of which was confirmed by surgical pathology. Patients were treated at three different hospitals. A training cohort of 180 patients was collected from the largest center, while an external validation cohort of 144 patients was collected from the other two centers. To extract radiomics features, regions of interest (ROIs) were outlined layer by layer along the edge of the tumor contour on CT images of the arterial and portal venous phases. The dimensionality of radiomic features was reduced, and the top 10 features with importance value above 5 were selected before modeling. The training cohort used three classifiers [logistic regression, support vector machine (SVM), and random forest] to establish three GIST risk stratification prediction models. The receiver operating characteristic curve (ROC) was used to compare model performance, which was validated by external data.Results: In the training cohort, the average area under the curve (AUC) was 0.84 ± 0.07 of the logistic regression, 0.88 ± 0.06 of the random forest, and 0.81 ± 0.08 of the SVM. In the external validation cohort, the AUC was 0.85 of the logistic regression, 0.90 of the random forest, and 0.80 of the SVM. The random forest model performed the best in both the training and the external validation cohorts and could be generalized.Conclusion: Based on CT radiomics, there are multiple machine-learning models that can predict the risk of GISTs. Among them, the random forest algorithm had the highest prediction efficiency and could be readily generalizable. Through external validation data, we assume that the random forest model may be used as an effective tool to guide preoperative clinical decision-making.

Machine Learning-Based Radiomics Nomogram for Detecting Extramural Venous Invasion in Rectal Cancer.

ObjectiveTo establish and validate a radiomics nomogram based on the features of the primary tumor for predicting preoperative pathological extramural venous invasion (EMVI) in rectal cancer using machine learning.MethodsThe clinical and imaging data of 281 patients with primary rectal cancer from April 2012 to May 2018 were retrospectively analyzed. All the patients were divided into a training set (n = 198) and a test set (n = 83) respectively. The radiomics features of the primary tumor were extracted from the enhanced computed tomography (CT), the T2-weighted imaging (T2WI) and the gadolinium contrast-enhanced T1-weighted imaging (CE-TIWI) of each patient. One optimal radiomics signature extracted from each modal image was generated by receiver operating characteristic (ROC) curve analysis after dimensionality reduction. Three kinds of models were constructed based on training set, including the clinical model (the optimal radiomics signature combining with the clinical features), the magnetic resonance imaging model (the optimal radiomics signature combining with the mrEMVI status) and the integrated model (the optimal radiomics signature combining with both the clinical features and the mrEMVI status). Finally, the optimal model was selected to create a radiomics nomogram. The performance of the nomogram to evaluate clinical efficacy was verified by ROC curves and decision curve analysis curves.ResultsThe radiomics signature constructed based on T2WI showed the best performance, with an AUC value of 0.717, a sensitivity of 0.742 and a specificity of 0.621. The radiomics nomogram had the highest prediction efficiency, of which the AUC was 0.863, the sensitivity was 0.774 and the specificity was 0.801.ConclusionThe radiomics nomogram had the highest efficiency in predicting EMVI. This may help patients choose the best treatment strategy and may strengthen personalized treatment methods to further optimize the treatment effect.

Prediction of tumor size in patients with invasive ductal carcinoma using FT-IR spectroscopy combined with chemometrics: a preliminary study.

Precise detection of tumor size is essential for early diagnosis, treatment, and evaluation of the prognosis of breast cancer. However, there are some errors between the tumor size of breast cancer measured by conventional imaging methods and the pathological tumor size. Invasive ductal carcinoma (IDC) is a common pathological type of breast cancer. In this study, serum Fourier transform infrared spectroscopy (FT-IR) combined with chemometric methods was used to predict the maximum diameter and maximum vertical diameter of tumors in IDC patients. Three models were evaluated based on the pathological tumor size measured after surgery and included grid search support vector machine regression (GS-SVR), back propagation neural network optimized by genetic algorithm (GA-BP-ANN), and back propagation neural network optimized by particle swarm optimization (PSO-BP-ANN). The results show that three models can accurately predict tumor size. The GA-BP-ANN model provided the best fitting quality of the largest tumor diameter with the determination coefficients of 0.984 in test set. And the GS-SVR model provided the best fitting quality of the largest vertical tumor diameter with the determination coefficients of 0.982 in test set. The GS-SVR model had the highest prediction efficiency and the lowest time complexity of the models. The results indicate that serum FT-IR spectroscopy combined with chemometric methods can predict tumor size in IDC patients. In addition, compared with traditional imaging methods, we found that the experimental results of the three models are better than traditional imaging methods in terms of correlation and fitting degree. And the average fitting error of PSO-BP-ANN and GA-BP-ANN models was less than 0.3mm. The minimally invasive detection method is expected to be developed into a new clinical diagnostic method for tumor size estimation to reduce the diagnostic trauma of patients and provide new diagnostic experience for patients. Graphical Abstract.

Identification of an individualized RNA binding protein-based prognostic signature for diffuse large B-cell lymphoma.

RNA binding proteins (RBPs) are increasingly appreciated as being essential for normal hematopoiesis and have a critical role in the progression of hematological malignancies. However, their functional consequences and clinical significance in diffuse large B‐cell lymphoma (DLBCL) remain unknown. Here, we conducted a systematic analysis to identify RBP‐related genes affecting DLBCL prognosis based on the Gene Expression Omnibus database. By univariate and multivariate Cox proportional hazards regression (CPHR) methods, six RBPs‐related genes (CMSS1, MAEL, THOC5, PSIP1, SNIP1, and ZCCHC7) were identified closely related to the overall survival (OS) of DLBCL patients. The RBPs signature could efficiently distinguished low‐risk from high‐risk patients and could serve as an independent and reliable factor for predicting OS. Moreover, Gene Set Enrichment Analysis revealed 17 significantly enriched pathways between high‐ versus low‐risk group, including the regulation of autophagy, chronic myeloid leukemia, NOTCH signaling pathway, and B cell receptor signaling pathway. Then we developed an RBP‐based nomogram combining other clinical risk factors. The receiver operating characteristic curve analysis demonstrated high prognostic predictive efficiency of this model with the area under the curve values were 0.820 and 0.780, respectively, in the primary set and entire set. In summary, our RBP‐based model could be a novel prognostic predictor and had the potential for developing treatment targets for DLBCL.

Identifying atmospheric pollutant sources using a machine learning dispersion model and Markov chain Monte Carlo methods

Estimating the sources of contaminant or hazard emissions is important for pollution control and safety management. Markov chain Monte Carlo (MCMC), combined with Bayesian inference, was used to identify the source terms of pollutants. However, the efficiency and accuracy of the forward dispersion model greatly impacted the performance of the estimation method. Therefore, a machine learning algorithm (MLA) model with high prediction accuracy and efficiency was proposed and coupled with MCMC method to estimate the source terms. A previously proposed MLA model was used to obtain the expected concentrations in Bayesian estimation. The Delayed Rejection Adaptive Metropolis (DRAM) method was applied to sample particles in order to form Markov chains. To evaluate the performance of the MCMC–MLA method, a Gaussian dispersion model was selected as the forward model. The performances of MCMC–MLA and MCMC–Gaussian models were then compared with release cases in Prairie Grass experiment and the results showed that the MCMC–MLA method converged more rapidly than the MCMC–Gaussian model. Nevertheless, release cases in the Round Hill experiment were also used to test the generalisability of the MCMC–MLA. The results indicated that the performance of MCMC–MLA was better than that of the MCMC–Gaussian model for estimating source terms in estimation accuracy. Hence, the MCMC–MLA method proposed here is potentially a useful tool for identifying emissions source parameters with high accuracy and efficiency, as well as reasonable probability estimates.

A new net primary productivity model and new management strategy of grassland classification based on CSCS in China

The discovery of grassland field, form and biomass in China was central to the sustainable development of grassland. In this study, the realistic spatial distribution patterns of grasslands were clarified through the combination of the International Geosphere-Biosphere Program (IGBP) and the Comprehensive and Sequential Classification System (CSCS). An optimal net primary productivity (NPP) model suitable for Chinese grasslands was introduced by integrating the classification indices-based model (CIM) with the Normalised Difference Vegetation Index (NDVI), and comparing it with the standard classical model (Miami, Schuur, CIM, CASA model). Using the optimal model as the algorithm basis, the net primary production spatial pattern of grassland in China was determined. The results showed that: (1) the total area of grassland was ~374.3 × 104 km2 in 2018, mainly distributed in north-western China. Among the grassland super-class groups, Tundra and alpine steppe were largest, and Warm desert smallest; (2) the optimal modified CIM had the highest prediction efficiency, and the overall accuracy was higher than the standard classical model (Miami, Schuur, CIM, CASA model). It achieved the accurate calculation of grassland NPP in China; (3) different grassland super-class groups had different carbon fixation efficiency per unit area, resulting in huge differences in total NPP. Among the various grassland super-class groups, the temperate humid grassland, steppe, tundra and alpine steppe had high conversion efficiency per unit area of NPP, whereas that for warm desert and the savanna was low. The total NPP was 388.04 × 1012 g C/year in the study area in 2018. The results provide a basis for the rational arrangement of grassland ecological and productive functions, and are significant for developing a new strategy of grassland classification management in China.

Artificial neural network models for predicting the corrosion behavior of friction stir processed AA5083

Aluminum alloy 5083 has good corrosion resistance in marine environments and consists of a primary phase of α-Al and a secondary phase of β-Mg2Al3. AA5083 was subjected to friction stir processing (FSP. AA5083 was friction stir processed by varying the tool rotation speed, tool traverse speed, and tool shoulder diameters as per face-centered central composite design. The specimens were subjected to a potentiodynamic polarization test in an artificial seawater solution to estimate the corrosion potential and corrosion rate of the alloy at three different temperatures. In this work, a feed-forward backpropagation network with the Levenberg–Marquardt training algorithm was developed to predict the corrosion potential and corrosion rate of the specimens as a function of the friction stir processing process parameters. The model predictions were in good agreement with the experimental results. The correlation coefficient of the models was approximately equal to unity, demonstrating the high prediction efficiency of the developed models.

Predicting the Risk of Depression in the Elderly by Immunological Indicators Research

Objective: to construct a mathematical model that predicts the state of depression by immunological parameters in the blood plasma of older people to further predict the development of the disease.Patients and methods: 55 hospitalized patients of late age (mean age 69.2 ± 6.9 years) with a depressive episode were included in the study. The control group consisted of 41 elderly people (average age 66.6 ± 6.2 years) without depressive disorders. The activity of inflammatory and autoimmune markers in the blood plasma of patients and control groups was determined: the enzymatic activity of leukocyte elastase (LE), the functional activity of the α1-proteinase inhibitor (α1-PI), the level of autoantibodies to neuro-specific antigens S100B and the myelin basic protein (MBP). Statistical data processing was performed using the R (R version 3.2.4) and STATA (version 12.1) programs. We used point-bead-correlation to measure the strength and direction of the relationship between the binary variable and continuous variables and logistic regression to predict the probability of occurrence of events of interest by the values of one or more independent variables (predictors).Results: in patients with depressive disorders, a statistically significant increase in the functional activity of α1-PI (p ≤ 0.05) and the level of autoantibodies to the neurospecific S100B antigen (p ≤ 0.05) was revealed compared with the control. LE activity and MBP level did not differ from the control (p = 0.12 and p = 0.1, respectively). Based on immunological parameters in elderly patients with depression, a mathematical model is constructed. The accuracy of the correct prediction of outcomes using the model as a whole was 83.33%, which indicates a high predictive efficiency of this model.Conclusion: the results of mathematical analysis obtained in this work indicate that immunological parameters such as the functional activity of α1-PI and S100B are statistically significantly associated with the likelihood of depression in the elderly. Indicators such as enzymatic activity of LE and the level of autoantibodies to MBP did not have a statistically significant effect on the desired probability.

A Prognostic Nomogram Model Based on mRNA Expression of DNA Methylation-Driven Genes for Gastric Cancer.

PurposeThe exploration and interpretation of DNA methylation-driven genes might contribute to molecular classification, prognostic prediction and therapeutic choice. In this study, we built a prognostic risk model via integrating analysis of the transcriptome and methylation profile for patients with gastric cancer (GC).MethodsThe mRNA expression profiles, DNA methylation profiles and corresponding clinicopathological information of 415 GC patients were downloaded from The Cancer Genome Atlas (TCGA). Differential expression and correlation analysis were performed to identify DNA methylation-driven genes. The candidate genes were selected by univariate Cox regression analyses followed by the least absolute shrinkage and selection operator (LASSO) regression. A prognostic risk nomogram model was then built together with clinicopathological parameters.Results5 DNA methylation-driven genes (CXCL3, F5, GNAI1, GAMT and GHR) were identified by integrated analyses and selected to construct the prognostic risk model with clinicopathological parameters. High expression and low DNA hypermethylation of F5, GNAI1, GAMT and GHR, as well as low expression and high DNA hypomethylation of CXCL3 were significantly associated with poor prognosis rates, respectively. The high-risk group showed a significantly shorter prognosis than the low-risk group in the TCGA dataset (HR = 0.212, 95% CI = 0.139–0.322, P = 2e-15). The final nomogram model showed high predictive efficiency and consistency in the training and validation group.ConclusionWe construct and validate a prognostic nomogram model for GC based on five DNA methylation-driven genes with high performance and stability. This nomogram model might be a powerful tool for prognosis evaluation in the clinic and also provided novel insights into the epigenetics in GC.

Predicted and measured hydraulic conductivity of sand-sized crushed limestone

The hydraulic conductivity of 54 sand-sized crushed limestone materials was measured by conducting constant head tests in a rigid-wall permeameter and was estimated using six predictive equations requiring easily obtainable parameters. The gradations tested had effective grain size, D10, from 0.079 to 2.15 mm; uniformity coefficient, Cu, from 1.19 to 15.79; and void ratio, e, from 0.42 to 0.76. The measured hydraulic conductivity had a range of about three orders of magnitude (3.4*10−3 to 3.3 cm/s). Four of the predictive equations, based on the square of the effective grain size, D102, yielded closely grouped results differing by not more than a factor of 2. Long existing equations by Terzaghi (1925) and by Hazen (1892), adjusted for void ratio according to Taylor (1948), were found to have a high predictive efficiency with a ratio of predicted to measured values between 1/2 and 2 for 70% of the materials tested. The Kenney et al. (1984) equation, based on D52, was also efficient but underestimated measured values for 63% of all cases. The Kozeny–Carman equation (Taylor 1948; Chapuis 2012), based on specific surface, overestimated measurements for 90% of the tested materials by a factor of up to 3.

Clariant relaunches campaign to sell its organic pigments business

Aluminum alloy 5083 has good corrosion resistance in marine environments and consists of a primary phase of α-Al and a secondary phase of β-Mg2Al3. AA5083 was subjected to friction stir processing (FSP. AA5083 was friction stir processed by varying the tool rotation speed, tool traverse speed, and tool shoulder diameters as per face-centered central composite design. The specimens were subjected to a potentiodynamic polarization test in an artificial seawater solution to estimate the corrosion potential and corrosion rate of the alloy at three different temperatures. In this work, a feed-forward backpropagation network with the Levenberg–Marquardt training algorithm was developed to predict the corrosion potential and corrosion rate of the specimens as a function of the friction stir processing process parameters. The model predictions were in good agreement with the experimental results. The correlation coefficient of the models was approximately equal to unity, demonstrating the high prediction efficiency of the developed models.

Driverless artificial intelligence framework for the identification of malignant pleural effusion

Our study aimed to explore the applicability of deep learning and machine learning techniques to distinguish MPE from BPE. We initially used a retrospective cohort with 726 PE patients to train and test the predictive performances of the driverless artificial intelligence (AI), and then stacked with a deep learning and five machine learning models, namely gradient boosting machine (GBM), extreme gradient boosting (XGBoost), extremely randomized trees (XRT), distributed random forest (DRF), and generalized linear models (GLM). Furthermore, a prospective cohort with 172 PE patients was applied to detect the external validity of the predictive models. The area under the curve (AUC) in the training, test and validation set were deep learning (0.995, 0.848, 0.917), GBM (0.981, 0.910, 0.951), XGBoost (0.933, 0.916, 0.935), XRT (0.927, 0.909, 0.963), DRF (0.906, 0.809, 0.969), and GLM (0.898, 0.866, 0.892), respectively. Although the Deep Learning model had the highest AUC in the training set (AUC = 0.995), GBM demonstrated stable and high predictive efficiency in three data sets. The final AI model by stacked ensemble yielded optimal diagnostic performance with AUC of 0.991, 0.912 and 0.953 in the training, test and validation sets, respectively. Using the driverless AI framework based on the routinely collected clinical data could significantly improve diagnostic performance in distinguishing MPE from BPE.