Artificial Neural Network Prediction Research Articles

The role of artificial neural networks in prediction of severe acute pancreatitis associated acute respiratory distress syndrome: A retrospective study.

Early identification and intervention of acute respiratory distress syndrome (ARDS) are particularly important. This study aimed to construct predictive models for ARDS following severe acute pancreatitis (SAP) by artificial neural networks and logistic regression. The artificial neural networks model was constructed using clinical data from 214 SAP patients. The patient cohort was randomly divided into a training set and a test set, with 149 patients allocated to the training set and 65 patients assigned to the test set. The artificial neural networks and logistic regression models were trained by the training set, and then the performance of both models was evaluated using the test set. The sensitivity, specificity, PPV, NPV, accuracy, and AUC value of artificial neural networks model were 68.0%, 87.5%, 77.3%, 81.4%, 80.0%, 0.853 ± 0.054 (95% CI: 0.749-0.958). The sensitivity, specificity, PPV, NPV, accuracy and AUC value of logistic regression model were 48.7%, 85.3%, 65.5%, 74.4%, 72.0%, 0.799 ± 0.045 (95% CI: 0.710-0.888). There were no significant differences between the artificial neural networks and logistic regression models in predictive performance. Bedside Index of Severity in Acute Pancreatitis score, procalcitonin, prothrombin time, and serum calcium were the most important predictive variables in the artificial neural networks model. The discrimination abilities of logistic regression and artificial neural networks models in predicting SAP-related ARDS were similar. It is advisable to choose the model according to the specific research purpose.

Artificial neural network prediction of the initial stiffness of semi-rigid beam-to-column connections

Joints are significant components in the design and construction of steel structures. The characteristic parameters of the connections must be reproduced in a reliable way to represent the actual behaviour of a structure. Accordingly, the study of semi-rigid joints is essential to better understand this issue. Among the different types of semi-rigid joints, angle connections stand out as a suitable solution in many cases. This paper presents a methodology using artificial neural networks for predicting the initial rotational stiffness of major axis symmetrical angle connections according to the Eurocode description. A consistent stiffness database was developed from the existing data in the Steel Connection Data Bank. Then, the database was cleansed to provide with a robust training set. Different network architectures were analysed until a topology that showed a good performance and generalisation features was obtained. The network was successfully checked with some saved tests from the database and with off-database tests; the network could be reliably used within the range of the training input parameters.

Artificial Neural Network prediction of forming limit diagram for directionally-rolled, size scaled copper strips

Estimating the forming limit diagram (FLD) is tedious and cost-intensive. Methods driven by data and artificial intelligence are used to determine the relationship between scaled thickness and the forming rates of various cups drawn out of ETP copper sheets. Machine learning (ML) techniques have a good chance of predicting the FLD of copper alloys, and they are being used increasingly in sensitive electronic and structural applications. The current research aims to create ML-based artificial neural network (ANN) tools to model the relationship between scaled thickness and forming rates as a function of formability. The FLD is measured for copper strips of initial dimensions of 1500 mm long, 750 mm wide, and 6 mm thick, whose thickness was reduced successively by 50% in nine incremental steps. Thus, 3, 1.5, 0.75, 0.38, and 0.19 mm sheets were obtained and used to determine FLD through the Nakajima approach. An FEA model of the drawing was made in Altair Inspire Form, and the simulation results were used to train a two-step ML. A Bayesian regularization (BR) and Levenberg-Marquardt algorithm (LM) were used in the first step to predict strains’ maximum and minimum points. In the second step, the minor strains predicted in the first step are used as inputs. Using the same feature set, the BR and LM algorithms predict the major strain, showing a linear trend until the middle and then a nonlinear trend. The trained ML model was used to predict unknown intermediate values for estimating the over-learning and over-fitting problems here for 2 and 0.25 mm thick sheets and are validated experimentally. The variation between the FLDs of predicted and experimentally verified data falls between 2% and 5%. Such small changes in the FLD values show that the proposed ML model could be used to predict the FLDs of copper strips.

Application of Constitutive Models and Machine Learning Models to Predict the Elevated Temperature Flow Behavior of TiAl Alloy.

The hot deformation behaviors of a Ti46Al2Cr2Nb alloy were investigated at strain rates of 0.001-0.1 s-1 and temperatures of 910-1060 °C. Under given deformation conditions, the activation energy of the TiAl alloy could be estimated as 319 kJ/mol. The experimental results were predicted by different predictive models including three constitutive models and three data-driven models. The most accurate data-driven model and constitutive model were an artificial neural network (ANN) and an Arrhenius type strain-compensated Sellars (SCS) model, respectively. In addition, the generalization capability of ANN model and SCS model was examined under different deformation conditions. Under known deformation conditions, the ANN model could accurately predict the flow stress of TiAl alloys at interpolated and extrapolated strains with a coefficient of determination (R2) greater than 0.98, while the R2 value of the SCS model was smaller than 0.5 at extrapolated strains. However, both ANN and SCS models performed poorly under new deformation conditions. A hybrid model based on the SCS model and ANN predictions was shown to have a wider generalization capability. The present work provides a comprehensive study on how to choose a predictive model for the flow stress of TiAl alloys under different conditions.

Structural assessment of Basalt FRP reinforced self-compacting geopolymer concrete using artificial neural network (ANN) modelling

Basalt FRP geopolymer composites are considered a sustainable solution to address the conventional OPC-Steel composite structures' structural and environmental concerns, especially the increased carbon emissions and structural deterioration due to corrosion attacks. Recent research on novel composites reports the superior performance of these FRP-geopolymer concrete structural applications reporting high strength and durability. However, the existing design guidelines do not accurately predict the flexural capacity and strength characteristics, which hinders the broader use of FRP-based geopolymer composites. This study aims to develop a prediction tool to help the researchers model the structural performance of the novel FRP-geopolymer composite based on modern machine learning algorithms. It shall address the lack of data on novel composites' structural and long-term performance in various exposure conditions. This paper thus reports the suitability of Artificial Neural Network (ANN) based models to accurately predict the flexural strength of Basalt FRP reinforced self-compacting geopolymer concrete beams under various exposure conditions. The ANN model is trained and validated against the experimental data from the long-term study conducted on the beam specimens subjected to one-year ambient and marine exposures. The ANN prediction model developed in this study reported close agreement with the experimental data showing better prediction accuracy than the existing empirical and numerical models.

Mix design of equal strength high volume fly ash concrete with artificial neural network

High volume fly ash concrete (HVFAC) has been widely used, and the mix proportion of HVFAC is usually designed based on the required compressive strength. Therefore, the research of HVFAC under the condition of equal strength has more engineering application value. However, in the current mix proportion design standard, there is no special discussion on the mix proportion design method of HVFAC. In this study, the compressive strength of HVFAC affected by four factors (water-binder ratio, FA content, curing method and curing age) was systematically studied. Based on 594 sets of strength data obtained from experiments, the method to determine the key mix proportion parameters (water-binder ratio and FA content) of equal strength HVFAC was proposed through machine learning algorithms (ML). Two ML methods (i.e., a multiple linear regression (MLR) and an artificial neural networks (ANN)) were developed to predict the HVFAC compressive strength and then their performances were compared. Through comparison, it was found that the mean absolute error and root mean squared error of the ANN model were both lower than the MLR model, and the ANN model has lower error and higher accuracy. In addition, the reliability of the proposed ANN model was verified with data in other literatures, the results showed the errors between the predicted values and the measured values are lower than 25%. The weight contribution rate of each factor to the strength was calculated. Among the four factors, the curing age had the greatest impact on the compressive strength, which contributes up to 60.3%. The influence degree of FA content is greater than that of water-binder ratio. With the development of curing age, the influence degree of FA content and water-binder ratio on the strength gradually decreases while the curing method increases. Finally, according to the ANN prediction results, the key mix parameters design method for equal strength HVFAC under multi-factor conditions was proposed, which provide guidance for concrete preparation in actual engineering.

Predictive modeling and analysis of key drivers of groundwater nitrate pollution based on machine learning

Nitrate comtamination of shallow groundwater in agricultural intensification regions is a prevalent and global environment issue affecting food security, human health, and the water ecology. Developing a prediction model for groundwater nitrate contamination is crucial for protection of groundwater resources. Machine learning modeling offer potentials to predict contamination, but often fails to adequately screen input features in complex non-linear environments before modeling. Additionally, there is a tendency to overlook the interpretable description of the relationship between major water chemistry parameters and nitrate concentration after modeling, which hinders scientific decision-making by water resource managers. In this study, a dataset consisting of hydrochemical test results from 316 groundwater samples collected between 2011 and 2015 in intensive agricultural areas of Northeast China was collected. Prior to modeling, the dataset was grouped based on land use type, vadose zones types, and thickness. Based on the grouping of data sets, self-organizing map (SOM) and Spearman's coefficients were employed to identity the correlations between water chemical parameters and nitrate concentration (NO3–-N), which provided a logical basis for selecting the key input variables. A radial basis function artificial neural network (RBF ANN) prediction model and principal components regression (PCR) models was constructed using the dataset, and particle swarm optimization algorithm was applied to determine optimal parameter combinations of the RBF ANN. After conducting modeling, the advantages and disadvantages of RBF ANN and principal components regression (PCR) models were thoroughly examined and discussed, and the primary factors that impact nitrate concentrations in groundwater were analyzed using a PCR model. The results revealed that the RBF ANN model showed greater accuracy, while the PCR model offered better interpretability. Therefore, the integration of the two models is advantageous for nitrate prediction research. Human activities, rainfall, land-use types, vadose zones types and thickness are external factors that affecting nitrate concentration, and the difference between nitrification. Denitrification, and dissimilatory nitrate reduction to ammonium (DNRA) is internal driving factors for the change of nitrate concentration in the aquifer, and partially determine differences in the spatial distribution of aquifer nitrate concentrations under various land use types. This study provides a comprehensive approach for predicting nitrate pollution in groundwater under non-linear environmental conditions, supporting decision making in sustainable groundwater development, management, and conservation.

Data augmentation driven by optimization for membrane separation process synthesis

This paper proposes a new hybrid strategy to optimally design membrane separation problems. We formulate the problem as a Non-Linear Programming (NLP) model. A common approach to represent the physical behavior of the membrane is to discretize the system of differential equations that govern the separation process. Instead, we represent the input/output behavior of the single membrane by an artificial neural network (ANN) predictor. The ANN is trained on a dataset obtained through the MEMSIC simulator. The equation form of the trained predictor (shape and weights) is then inserted in the NLP model at the place of the discretized system of differential equations.To improve the ANN accuracy without an excessive computational burden, we propose data augmentation strategies to target the regions where densify the dataset. We compare a data augmentation strategy from the literature with a novel one that densifies the dataset around the stationary points visited by a global optimization algorithm.Our approach was validated using a relevant industrial case study: hydrogen purification. Validation by simulation is performed on the obtained solutions. The computational results show that a data augmentation smartly coupled with optimization can produce a robust and reliable design tool.

Determining the influence of meteorological parameters on outdoor thermal comfort using ANFIS and ANN

The study aims to develop artificial neural networks for prediction of outdoor thermal comfort using meteorological parameters as input parameters. Universal Thermal Climate Index (UTCI) is used as the target parameter. For this purpose, a total number of 5088 hours of field monitoring data was considered from four representative urban streets of Gwalior city, India. First, linear association was determined between meteorological parameters. Mean radiant temperature was to be in high correlation with globe temperature and surface temperature. Second, Adaptive Neuro Fuzzy Inference System (ANFIS) was used to rank the meteorological parameters in order of their impact on UTCI. Air temperature was found to be having highest influence. Third, ANN models are developed to predict UTCI with air temperature as the only meteorological parameter in input layer. The developed ANN models for all four streets show remarkable predictive ability for both summer (R2 = 0.852, 0.986, 0.962, 0.955) and winter season (R2 = 0.976, 0.870, 0.941, 0.950). Additionally, the success index of the developed models is found to be in range 0.73 – 1, 0.88 – 1, 0.86 – 1, 0.87 – 1 for summer season and 0.78 – 0.99, 0.61 – 0.98, 0.55 – 0.98, 0.87 – 0.99 for winter season. The study contributes to the smart city initiatives for future urban designing by establishing that outdoor thermal comfort can be easily predicted using air temperature when other microclimatic parameters are difficult to record using machine learning approach.

The application of deep neural networks for the prediction of California Bearing Ratio of road subgrade soil

California Bearing Ratio (CBR) of the subgrade soil is one of the essential values for the design and construction of the asphalt pavement of highway projects. However, the estimation of the CBR value through laboratory tests is time-consuming and labor-intensive. Thus, this study focuses on providing artificial neural network (ANN) prediction models that can be efficiently used for the prediction of the CBR value of the subgrade soil in Egypt from the grain size distribution, Atterberg limits, and compaction parameters. 240 ANNs with different hyperparameters are investigated in order to optimize the hyperparameters so that the final chosen ANN can provide accurate results. The analysis shows that the deep neural networks outperform the shallow ANNs. Finally, comparing the performance of the ANNs with the traditional multiple linear regression (MLR) shows that ANNs outperform the MLR models as the ANNs have much better performance and can generate highly accurate predictions.

Implementation of Ant Colony Optimization – Artificial Neural Network in Predicting the Activity of Indenopyrazole Derivative as Anti-Cancer Agent

Cancer is a disease induced by the abnormal growth of cells in body tissues. This disease is commonly treated by chemotherapy. However, at first, cancer cells can respond to the activity of chemotherapy over time, but over time, resistance to cancer cells appears. Therefore, it is required to develop new anti-cancer drugs. Indenopyrazole and its derivative have been investigated to be a potential drug to treat cancer. This study aims to predict indenopyrazole derivative compounds as anti-cancer drugs by using Ant Colony Optimization (ACO) and Artificial Neural Network (ANN) methods. We used 93 compounds of indenopyrazole derivative with a total of 1876 descriptors. Then, the descriptors were reduced by using the Pearson Correlation Coefficient (PCC) and followed by the ACO algorithm to get the most relevant features. We found that the best number of descriptors obtained from ACO is ten descriptors. The ANN prediction model was developed with three architectures, which are different in hidden layer number, i.e., 1, 2, and 3 hidden layers. Based on the results, we found that the model with three hidden layers gives the best performance, with the value of the R2 test, R2 train, and Q2 train being 0.8822, 0.8495, and 0.8472, respectively.

Predicting geometric errors and failures in additive manufacturing

PurposeObjects fabricated using additive manufacturing (AM) technologies often suffer from dimensional accuracy issues and other part-specific problems. This study aims to present a framework for estimating the printability of a computer-aided design (CAD) model that expresses the probability that the model is fabricated correctly via an AM technology for a specific application.Design/methodology/approachThis study predicts the dimensional deviations of the manufactured object per vertex and per part using a machine learning approach. The input to the error prediction artificial neural network (ANN) is per vertex information extracted from the mesh of the model to be manufactured. The output of the ANN is the estimated average per vertex error for the fabricated object. This error is then used along with other global and per part information in a framework for estimating the printability of the model, that is, the probability of being fabricated correctly on a certain AM technology, for a specific application domain.FindingsA thorough experimental evaluation was conducted on binder jetting technology for both the error prediction approach and the printability estimation framework.Originality/valueThis study presents a method for predicting dimensional errors with high accuracy and a completely novel approach for estimating the probability of a CAD model to be fabricated without significant failures or errors that make it inappropriate for a specific application.

Predviđanje adhezivne čvrstoće nekih lakova uz pomoć modela mekog računalstva

The purpose of this study was to predict the adhesion strength of the varnish, which is applied as a protective coating/finish on the surface of wooden material using soft computing models. In this study, the soft computing approaches were applied to oak (Quercus Petrea L.), chestnut (Castanea sativa M.), and scotch pine (Pinus sylvestris L.) with water-based, polyurethane, and acrylic varnishes. The adhesion strength of the varnish was determined in accordance with the Turkish Standard Institute-24624 and ASTM D4541. The outcome of the experiment was used to develop artificial neural network (ANN) and fuzzy logic (FL) prediction models. The total number of 360 data points was split as 80 % of training and 20 % of test for the model development. During the application of the ANN, 6 features were used as an input, while the adhesion strength was used as an output of the model. The coefficient of determination values (R2) for training and testing in the ANN models were 0.9939 and 0.9580, respectively. In the case of the ANFIS model, R2 values for training and testing were 0.9917 and 0.9929, respectively. Considering the MAPE, RMSE, and R2 values obtained from the results of both training and test values, it can be concluded that the ANFIS model showed a more successful performance in estimating varnish adhesion strength. Therefore, ANN and ANFIS have the potential to provide time and cost-efficient benefits in estimating wood adhesion strength.

Portfolio Optimization Based on Return Prediction and Semi Absolute Deviation (SAD)

A portfolio is a collection of investment financial assets managed by financial institutions or individuals. In investment activities, investors expect minimal loss risk and optimal stock portfolio weight to get maximum profit. Investors can monitor changes in stock index values to compare portfolio performance. This research has discussed how to build a portfolio based on stock datasets with the LQ45 index using return predictions from the artificial neural network (ANN) method with semi-absolute deviation (SAD). Furthermore, the portfolio is optimized by looking for weights that match it. After that, a comparison of portfolio performance was carried out using the Sharpe ratio (SR) method between the semi-absolute deviation (SAD) portfolio and the portfolio resulting from the formation of the equal weight (EW) portfolio. Portfolio performance with ANN prediction and SAD is better than equal-weight portfolios in terms of mean return, standard deviation, and sharpe ratio for portfolios with few stocks, namely 2 and 3 stocks. In addition, a portfolio with a higher number of stocks can make the portfolio value from the ANN close prediction algorithm process and the selection of weights based on SAD is better than portfolios with equal weight for each list of stocks in the portfolio.

Zoning Prediction and Mapping of Three-Dimensional Forest Soil Organic Carbon: A Case Study of Subtropical Forests in Southern China

Accurate soil organic carbon (SOC) maps are helpful for guiding forestry production and management. Different ecological landscape areas within a large region may have different soil–landscape relationships, so models specifically for these areas may capture these relationships more accurately than the global model for the entire study area. The aim of this study was to investigate the role of zonal modelling in predicting forest SOC and to produce highly accurate forest SOC distribution maps. The prediction objects were SOC at five soil depths (0–20, 20–40, 40–60, 60–80, and 80–100 cm). First, the forest type map and soil texture class map were used to divide the relative homogeneous regions in Shaoguan City, Guangdong Province, China. Second, seven terrain variables derived from a 12.5-m digital elevation model (DEM) and five vegetation variables generated from 10-m Sentinel-2 remote sensing images were used as predictors to develop regional artificial neural network (ANN) models for each homogeneous region, as well as a global ANN model for the entire study area (1000 sample points). Finally, 10-fold cross-validation was used to assess the ANN prediction model performance, and independent validation was used to evaluate the produced forest SOC prediction maps (194 additional samples). The cross-validation results showed that the accuracies of the regional models were better than that of the global model. Independent validation results also showed that the precision (R2) of 0- to 100-cm forest SOC maps generated by forest type modelling had an improvement of 0.05–0.15, and that by soil texture class modelling had an improvement of 0.07–0.13 compared to the map generated by the global model. In conclusion, delineating relatively homogeneous regions via simple methods can improve prediction accuracy when undertaking soil predictions over large areas, especially with complex forest landscapes. In addition, SOC in the study area is generally more abundant in broadleaf forest and clay areas, with overall levels decreasing with soil depth. Accurate SOC distribution information can provide references for fertilization and planting. Plants with particularly high soil fertility requirements may perhaps be planted in broadleaf forests or clay areas, and plants with particularly developed roots may require furrow application of a small amount of SOC.

Differential diagnosis of lung cancer and tuberculosis based on 18 F-fluorodeoxyglucose PET/CT multi-time points imaging.

To investigate the value of 18 F-fluorodeoxyglucose(FDG) PET/CT multi-time points imaging (MTPI) on the differential diagnosis between lung cancer (LC) and tuberculosis (TB). Sixty-four patients underwent 18 F-FDG PET/CT MTPI. The stdSUVmax, stdSUVavg, retention index, metabolic tumor volume, total lesion glycolysis at four-time points and slope of metabolic curve were measured and calculated, and the sex, age, and uniformity of FDG uptake were recorded. The difference in each index between LC and TB was analyzed, and dynamic metabolic curves (DMCs) of LC and TB were fitted by significance indexes. Artificial neural network (ANN) prediction models were established between squamous cell carcinoma (SCC) and TB, as well as between adenocarcinomas and TB. Differences between SCC and TB, stdSUVmax/avg at four-time points, total lesion glycolysis, stdSUVmax/avg slope (1-2 h,1-3 h and 1-4 h), uniformity of FDG uptake and age were significant. stdSUVavg has the largest area under the 4 h curve; age was only significant between adenocarcinomas and TB. DMCs at 1-4 h fitted by stdSUVavg were more helpful in differentiating LC and TB than stdSUVmax. stdSUVavg(1 h and 4 h), stdSUVavg slope 1-4 h, age, and uniformity of FDG uptake were selected to establish an ANN prediction model between SCC and TB; the area under the curve (AUC) was 100.0%. The same indices were used to establish the prediction model between adenocarcinomas and TB; the AUC was up to 83.5, and after adding stdSUVavg (2 and 4 h) to adenocarcinomas and TB models, the AUC was 87.7%. 18 F-FDG PET/CT MTPI fitting DMCs and establishing an ANN prediction model would distinguish SCC from TB relatively accurately and provide certain help in the differentiation between adenocarcinomas and TB.

Measuring and Evaluating the Speed and the Physical Characteristics of Fishes Based on Video Processing

Acquiring the morphological parameters of fish with the traditional method (depending on human and non-automatic factors) not only causes serious problems, such as disease transmission, mortality due to stress, and carelessness and error, but it is also time-consuming and has low efficiency. In this paper, the speed of fish and their physical characteristics (maximum and minimum diameter, equivalent diameter, center of surface, and velocity of fish) were investigated by using a programmed online video-recording system. At first, using the spatial coordinates obtained from YOLOv2, the speed of the fish was calculated, and the morphological characteristics of the fish were also recorded using this program during two stages of feeding and normal conditions (when the fish are not in feeding condition). Statistical analysis was performed between the measured parameters due to the high correlation between the parameters, and the classification system with high accuracy was able to provide an accurate prediction of the fish in both normal and feeding conditions. In the next step, an artificial neural network (ANN) prediction model (with three neurons; four input, one hidden layer, and one output) was presented to plan the system online. The model has the lowest error (1.4 and 0.14, respectively) and the highest coefficient of explanation (0.95 and 0.94, respectively) in two modes, normal and feeding, which are presented by the ANN system for planning the online system. The high accuracy and low error of the system, in addition to having a high efficiency for continuous and online monitoring of live fish, can have a high economic benefit for fish breeders due to the simplicity of its equipment, and it can also check and diagnose the condition of fish in time and prevent economic damage.

Using Artificial Neural Networks for Predicting Ship Fuel Consumption

Abstract In marine vessel operations, fuel costs are major operating costs which affect the overall profitability of the maritime transport industry. The effective enhancement of using ship fuel will increase ship operation efficiency. Since ship fuel consumption depends on different factors, such as weather, cruising condition, cargo load, and engine condition, it is difficult to assess the fuel consumption pattern for various types of ships. Most traditional statistical methods do not consider these factors when predicting marine vessel fuel consumption. With technological development, different statistical models have been developed for estimating fuel consumption patterns based on ship data. Artificial Neural Networks (ANN) are some of the most effective artificial methods for modelling and validating marine vessel fuel consumption. The application of ANN in maritime transport improves the accuracy of the regression models developed for analysing interactive relationships between various factors. The present review sheds light on consolidating the works carried out in predicting ship fuel consumption using ANN, with an emphasis on topics such as ANN structure, application and prediction algorithms. Future research directions are also proposed and the present review can be a benchmark for mathematical modelling of ship fuel consumption using ANN.

On-the-fly accuracy evaluation of artificial neural networks and hybrid method to improve the robustness of neural network accelerated chemistry solving

A chemistry acceleration strategy based on the coupling of artificial neural networks (ANNs) and direct integration (DI) is proposed and evaluated in the context of turbulent combustion. The main novelty of this study is its focus on ANNs robustness assessment. A hybrid DI/ANN strategy is proposed, which allows for a direct control of the prediction errors. This control is achieved by evaluating the ANN prediction error after each inference. To this end, a simple yet novel criterion based on mass conservation is proposed and compared to a criterion based on the distance between the inferred state and the training database, as done previously in the literature in the context of on-the-fly learning. A two-dimensional turbulent premixed H2 ignition case is used to assess the performance of the strategy and challenge the two criteria. An a priori study demonstrates that the state space-based criterion cannot correctly describe the ANN error, while the mass conservation-based one provides a good match with the ANN error. An a posteriori evaluation, involving actual simulations of the turbulent case, shows the ability of the hybrid DI/ANN model based on mass conservation error to improve the quality of the predictions and thus the robustness of ANNs. The increase in computational cost due to the hybrid model is acceptable as the DI is only applied in very localized regions in space and time.

Development and validation of an artificial neural network prediction model for postpartum hemorrhage with placenta previa.

Postpartum hemorrhage (PPH) is a leading cause of maternal morbidity worldwide and placenta previa is one of the major risk factors for PPH in overall population. However, the clinical prediction of PPH remains challenging. This study aimed to investigate an ideal machine learning-based prediction model for PPH in placenta previa parturients with cesarean section. The clinical data of 223 placenta previa parturients who underwent cesarean delivery in our hospital from 2016 to 2019 were retrospectively collected for analysis. An artificial neural network model was designed to predict PPH, defined as blood loss exceeding 1000 mL with 24h after delivery. Twenty clinical variables were selected as predictors. We also applied six conventional machine learning methods as reference models, including support vector machine, decision tree, random forest, gradient boosting decision tree, adaboost and logistic regression. All the models were validated using 5-fold cross-validation. The area under the receiver operating characteristic curve (AUC), precision, recall and the prediction accuracy of each model were reported. A total of 223 pregnant women were enrolled in this study, including 101 cases (45.29%) experienced PPH. The proposed model achieved superior prediction performance with an AUC of 0.917, an accuracy of 0.851, a precision score of 0.829 and a recall score of 0.851, which outperformed other six conventional machine learning methods. Compared to the conventional machine learning approaches, artificial neural network model shows discriminative ability in identifying women's risk of PPH with placenta previa during cesarean section.