Bayesian Regularization Neural Network Research Articles

ICA-LightGBM Algorithm for Predicting Compressive Strength of Geo-Polymer Concrete

The main goal of the present study is to investigate the capability of hybridizing the imperialist competitive algorithm (ICA) with an intelligent, robust, and data-driven technique named the light gradient boosting machine (LightGBM) to estimate the compressive strength of geo-polymer concrete (CSGCo). The hyper-parameters of the LightGBM algorithm have been optimized based on ICA and its accuracy improved. The obtained results from the proposed hybrid ICA-LightGBM are compared with the traditional LightGBM model as well as four different topologies of artificial neural networks (ANN) comprising a multi-layer perceptron neural network (MLP), radial basis function (RBF), generalized feed-forward neural network (GFFNN), and Bayesian regularized neural network (BRNN). The results of these models were compared based on three evaluation indices of R2, RMSE, and VAF for providing an objective evaluation of the performance and capability of the predictive models. Concerning the outcomes, the ICA-LightGBM with the R2 of (0.9871 and 0.9805), RMSE of (0.4703 and 1.3137), and VAF of (98.5773 and 98.0397) for training and testing phases, respectively, was a superior predictor to estimate the CSGCo compared to the LightGBM with the R2 of (0.9488 and 0.9478), RMSE of (0.9532 and 2.1631), and VAF of (94.3613 and 94.5173); the MLP with the R2 of (0.9067 and 0.8959), RMSE of (1.3093 and 3.3648), and VAF of (88.9888 and 84.9125); the RBF with the R2 of (0.8694 and 0.8055), RMSE of (1.4703 and 5.0309), and VAF of (86.3122 and 66.1888); the BRNN with the R2 of (0.9212 and 0.9107), RMSE of (1.1510 and 2.6569), and VAF of (91.4168 and 90.5854); and the GFFNN with the R2 of (0.9144 and 0.8925), RMSE of (1.1525 and 2.9415), and VAF of (91.4092 and 88.9088). Hence, the proposed ICA-LightGBM algorithm can be efficiently used in anticipating the CSGCo.

Novel design of intelligent Bayesian networks to study the impact of magnetic field and Joule heating in hybrid nanomaterial flow with applications in medications for blood circulation

In current study, the effects of magnetic field and Joule heating in flow of (Cu+CuO) hybrid nanomaterial model with blood as a base fluid is analyzed effectively by exploiting novel design of intelligent Bayesian regularization neural networks (IBRNN) for several important physical aspects. Mathematical models of the systems are derived by considering the small Eckert and Prandtl numbers with thermo-physical characteristics of Cu+CuO/blood hybrid nanoliquids while the heat emission/absorption is added in energy equation to maintain the temperature for blood flow. The effects of various parameters of interest like wall deformation rate, Joule heating parameter, magnetic parameter, volume fraction and permeability parameter on velocity and temperature distribution have been studied using viable numerical approach of Adams method as well as presented intelligent Bayesian networks. The reliability of the Bayesian networks is endorsed by convergence plots of MSEs, effective outputs of adaptive controlling parameters of optimization algorithm, error based histograms, and regression statistics for exhaustive simulations studies in case of sundry scenarios.

Comparing Algorithms for Estimation of Aboveground Biomass in Pinus yunnanensis

Comparing algorithms are crucial for enhancing the accuracy of remote sensing estimations of forest biomass in regions with high heterogeneity. Herein, Sentinel 2A, Sentinel 1A, Landsat 8 OLI, and Digital Elevation Model (DEM) were selected as data sources. A total of 12 algorithms, including 7 types of learners, were utilized for estimating the aboveground biomass (AGB) of Pinus yunnanensis forest. The results showed that: (1) The optimal algorithm (Extreme Gradient Boosting, XGBoost) was selected as the meta-model (referred to as XGBoost-stacking) of the stacking ensemble algorithm, which integrated 11 other algorithms. The R2 value was improved by 0.12 up to 0.61, and RMSE was decreased by 4.53 Mg/ha down to 39.34 Mg/ha compared to the XGBoost. All algorithms consistently showed severe underestimation of AGB in the Pinus yunnanensis forest of Yunnan Province when AGB exceeded 100 Mg/ha. (2) XGBoost-Stacking, XGBoost, BRNN (Bayesian Regularized Neural Network), RF (Random Forest), and QRF (Quantile Random Forest) have good sensitivity to forest AGB. QRNN (Quantile Regression Neural Network), GP (Gaussian Process), and EN (Elastic Network) have more outlier data and their robustness was poor. SVM-RBF (Radial Basis Function Kernel Support Vector Machine), k-NN (K Nearest Neighbors), and SGB (Stochastic Gradient Boosting) algorithms have good robustness, but their sensitivity was poor, and QRF algorithms and BRNN algorithm can estimate low values with higher accuracy. In conclusion, the XGBoost-stacking, XGBoost, and BRNN algorithms have shown promising application prospects in remote sensing estimation of forest biomass. This study could provide a reference for selecting the suitable algorithm for forest AGB estimation.

Estimating the Aboveground Biomass of Various Forest Types with High Heterogeneity at the Provincial Scale Based on Multi-Source Data

It is important to improve the accuracy of models estimating aboveground biomass (AGB) in large areas with complex geography and high forest heterogeneity. In this study, k-nearest neighbors (k-NN), gradient boosting machine (GBM), random forest (RF), quantile random forest (QRF), regularized random forest (RRF), and Bayesian regularization neural network (BRNN) machine learning algorithms were constructed to estimate the AGB of four forest types based on environmental factors and the variables selected by the Boruta algorithm in Yunnan Province and using integrated Landsat 8 OLI and Sentinel 2A images. The results showed that (1) DEM was the most important variable for estimating the AGB of coniferous forests, evergreen broadleaved forests, deciduous broadleaved forests, and mixed forests; while the vegetation index was the most important variable for estimating deciduous broadleaved forests, the climatic factors had a higher variable importance for estimating coniferous and mixed forests, and texture features and vegetation index had a higher variable importance for estimating evergreen broadleaved forests. (2) In terms of specific model performance for the four forest types, RRF was the best model both in estimating the AGB of coniferous forests and mixed forests; the R2 and RMSE for coniferous forests were 0.63 and 43.23 Mg ha−1, respectively, and the R2 and RMSE for mixed forests were 0.56 and 47.79 Mg ha−1, respectively. BRNN performed the best in estimating the AGB of evergreen broadleaved forests; the R2 was 0.53 and the RMSE was 68.16 Mg ha−1. QRF was the best in estimating the AGB of deciduous broadleaved forests, with R2 of 0.43 and RMSE of 45.09 Mg ha−1. (3) RRF was the best model for the four forest types according to the mean values, with R2 and RMSE of 0.503 and 52.335 Mg ha−1, respectively. In conclusion, different variables and suitable models should be considered when estimating the AGB of different forest types. This study could provide a reference for the estimation of forest AGB based on remote sensing in complex terrain areas with a high degree of forest heterogeneity.

Postharvest Treatments of Hass Avocado (Persea americana Mill.) and Estimation of Its Quality Using Hyperspectral Imaging (HSI)

Avocados’ shelf life is limited and difficult to monitor. This study evaluated the performance of chitosan coatings (1.5 and 2% w/v, T1 and T2) on avocados’ quality and shelf life against samples untreated (C) and treated with an ethylene inhibitor (1-MCP, M). Hyperspectral imaging (HSI) coupled with machine learning (ML) techniques was also evaluated to estimate Hass avocados’ quality indicators. Sensorial, physicochemical, and metabolic characteristics were measured using standard procedures. While T2 samples exhibited undesirable changes (i.e., uneven color and heterogeneous firmness), T1 behaved similarly to C. However, neither treatment could delay senescence as much as 1-MCP (42 vs ≤ 33 days). In general, Bayesian regularization neural networks (BRNNs) outperformed the other tested ML techniques in estimating quality attributes from HSI features, allowing for real-time nondestructive assessment of food quality. Adverse effects of chitosan coatings on avocados’ physiology were identified, which can inform the development of films with improved performance.

Imaging Sensor-Based High-Throughput Measurement of Biomass Using Machine Learning Models in Rice

Phenomics technologies have advanced rapidly in the recent past for precision phenotyping of diverse crop plants. High-throughput phenotyping using imaging sensors has been proven to fetch more informative data from a large population of genotypes than the traditional destructive phenotyping methodologies. It provides accurate, high-dimensional phenome-wide big data at an ultra-super spatial and temporal resolution. Biomass is an important plant phenotypic trait that can reflect the agronomic performance of crop plants in terms of growth and yield. Several image-derived features such as area, projected shoot area, projected shoot area with height constant, estimated bio-volume, etc., and machine learning models (single or multivariate analysis) are reported in the literature for use in the non-invasive prediction of biomass in diverse crop plants. However, no studies have reported the best suitable image-derived features for accurate biomass prediction, particularly for fully grown rice plants (70DAS). In this present study, we analyzed a subset of rice recombinant inbred lines (RILs) which were developed from a cross between rice varieties BVD109 × IR20 and grown in sufficient (control) and deficient soil nitrogen (N stress) conditions. Images of plants were acquired using three different sensors (RGB, IR, and NIR) just before destructive plant sampling for the quantitative estimation of fresh (FW) and dry weight (DW). A total of 67 image-derived traits were extracted and classified into four groups, viz., geometric-, color-, IR- and NIR-related traits. We identified a multimodal trait feature, the ratio of PSA and NIR grey intensity as estimated from RGB and NIR sensors, as a novel trait for predicting biomass in rice. Among the 16 machine learning models tested for predicting biomass, the Bayesian regularized neural network (BRNN) model showed the maximum predictive power (R2 = 0.96 and 0.95 for FW and DW of biomass, respectively) with the lowest prediction error (RMSE and bias value) in both control and N stress environments. Thus, biomass can be accurately predicted by measuring novel image-based parameters and neural network-based machine learning models in rice.

Novel computational protocol to support transfemoral prosthetic alignment procedure using machine learning techniques

BackgroundThe prosthetic alignment procedure considers biomechanical, anatomical and comfort characteristics of the amputee to achieve an acceptable gait. Prosthetic malalignment induces long-term disease. The assessment of alignment is highly variable and subjective to the experience of the prosthetist, so the use of machine learning could assist the prosthetist during the judgment of optimal alignment. Research objectiveTo assist the prosthetist during the assessment of prosthetic alignment using a new computational protocol based on machine learning. MethodsSixteen transfemoral amputees were recruited for training and validation of the alignment protocol. Four misalignments and one nominal alignment were performed. Eleven prosthetic limb ground reaction force parameters were recorded. A support vector machine with a Gaussian kernel radial basis function and a Bayesian regularization neural network were trained to predict the alignment condition, as well as the magnitude and angle of required to align the prosthesis correctly. The alignment protocol was validated by one junior and one senior prosthetist during the prosthetic alignment of two transfemoral amputees. ResultsThe support vector machine-based model detected the nominal alignment 92.6 % of the time. The neural network recovered 94.11 % of the angles needed to correct the prosthetic misalignment with a fitting error of 0.51°. During the validation of the alignment protocol, the computational models and the prosthetists agreed on the alignment assessment. The gait quality evaluated by the prosthetists reached a satisfaction level of 8/10 for the first amputee and 9.6/10 for the second amputee. ImportanceThe new computational prosthetic alignment protocol is a tool that helps the prosthetist during the prosthetic alignment procedure thereby decreasing the likelihood of gait deviations and musculoskeletal diseases associated with misalignments and consequently improving the amputees-prosthesis adherence.

Comparing Global Sentinel-2 Land Cover Maps for Regional Species Distribution Modeling

Mapping the spatial and temporal dynamics of species distributions is necessary for biodiversity conservation land-use planning decisions. Recent advances in remote sensing and machine learning have allowed for high-resolution species distribution modeling that can inform landscape-level decision-making. Here we compare the performance of three popular Sentinel-2 (10-m) land cover maps, including dynamic world (DW), European land cover (ELC10), and world cover (WC), in predicting wild bee species richness over southern Norway. The proportion of grassland habitat within 250 m (derived from the land cover maps), along with temperature and distance to sandy soils, were used as predictors in both Bayesian regularized neural network and random forest models. Models using grassland habitat from DW performed best (RMSE = 2.8 ± 0.03; average ± standard deviation across models), followed by ELC10 (RMSE = 2.85 ± 0.03) and WC (RMSE = 2.87 ± 0.02). All satellite-derived maps outperformed a manually mapped Norwegian land cover dataset called AR5 (RMSE = 3.02 ± 0.02). When validating the model predictions of bee species richness against citizen science data on solitary bee occurrences using generalized linear models, we found that ELC10 performed best (AIC = 2278 ± 4), followed by WC (AIC = 2367 ± 3), and DW (AIC = 2376 ± 3). While the differences in RMSE we observed between models were small, they may be significant when such models are used to prioritize grassland patches within a landscape for conservation subsidies or management policies. Partial dependencies in our models showed that increasing the proportion of grassland habitat is positively associated with wild bee species richness, thereby justifying bee conservation schemes that aim to enhance semi-natural grassland habitat. Our results confirm the utility of satellite-derived land cover maps in supporting high-resolution species distribution modeling and suggest there is scope to monitor changes in species distributions over time given the dense time series provided by products such as DW.

Photovoltaic Power Prediction Based on VMD-BRNN-TSP

Overfitting often occurs in neural network training, and neural networks with higher generalization ability are less prone to this phenomenon. Aiming at the problem that the generalization ability of photovoltaic (PV) power prediction model is insufficient, a PV power time-sharing prediction (TSP) model combining variational mode decomposition (VMD) and Bayesian regularization neural network (BRNN) is proposed. Firstly, the meteorological sequences related to the output power are selected by mutual information (MI) analysis. Secondly, VMD processing is performed on the filtered sequences, which is aimed at reducing the non-stationarity of the data; then, normalized cross-correlation (NCC) and signal-to-noise ratio (SNR) between the components obtained by signal decomposition and the original data are calculated, after which the key influencing factors are screened out to eliminate the correlation and redundancy of the data. Finally, the filtered meteorological sequences are divided into two datasets based on whether the irradiance of the day is zero or not. Meanwhile, the predictions are performed using BRNN for each of the two datasets. Then, the results are reordered in chronological order, and the prediction of PV power is realized conclusively. It was experimentally verified that the mean absolute value error (MAE) of the method proposed in this paper is 0.1281, which is reduced by 40.28% compared with the back propagation neural network (BPNN) model on the same dataset, the mean squared error (MSE) is 0.0962, and the coefficient of determination (R2) is 0.9907. Other error indicators also confirm that VMD is of much significance and TSP is contributive.

Landslide susceptibility prediction improvements based on a semi-integrated supervised machine learning model.

Differences in model application effectiveness, insufficient numbers of disaster samples, and unreasonable selection of non-hazard samples are common problems in landslide susceptibility studies. Therefore, in this paper, we propose a semi-integrated supervised approach to improve the prediction performance of machine learning (ML) models in landslide susceptibility studies. First, taking the lower reaches of the Jinsha River as the study area, a geospatial dataset consisting of 349 landslides, an equal number of randomly selected non-landslide points, and 12 environmental factors were randomly divided into training (70%) and testing (30%) datasets. Then, K-nearest neighbors (KNN), random forest (RF), and Bayesian-regularized neural network (BRNN) models were built. Second, the three models were combined to form an integrated weighted model. Very high- and low-prone areas were selected and, combined with the prediction results and remote sensing images, landslide and non-landslide samples were identified. The identified samples were then combined with the original samples to form new samples, which were used to sequentially construct the ensemble-supervised K-nearest neighbors (ESKNN), ensemble-supervised random forest (ESRF), and ensemble-supervised Bayesian-regularized neural network (ESBRNN) models. Finally, the area under the curve (AUC), true skill statistic (TSS), and frequency ratio (FR) values were used to test the accuracy of each model. The traditional ML model results and accuracy were improved by the semi-integrated supervised method. The ESRF model had the best prediction effect (AUC = 0.939, TSS = 0.440, and FR = 95.8%). The proposed semi-integrated supervised ML model solved the problems observed in traditional landslide susceptibility studies and provided insights for reducing variations in model applications, expanding landslide data sources, and improving non-landslide sample selection.

Final Fattening Live Weight Prediction in Anatolian Merinos Lambs from Some Body Characteristics at the Initial of Fattening by Using Some Data Mining Algorithms

This study's objective was to compare the performances of Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Bayesian Regularization Neural Network (BRNN) algorithms, which are some data mining algorithms used in final fattening live weight prediction. As the independent variable in the design of the algorithms, some body characteristics taken before fattening of 54 heads of Anatolian Merino lambs, with single birth and male, were withers height (WH), rump height (RH), body length (BL), chest girth (CG), leg girth (LG), and chest depth (CD) was used. The mean±standart errors for the body characteristics of Anatolian Merino lambs were determined to be 63.481±0.538, 63.315±0.501, 78.930±1.140, 60.037±0.549, 47.704±0.543, and 29.926±0.377, respectively. The mean initial live weight (ILW) and the mean final live weight (FLW) were found as 35.89±0.84 and 49.49±0.88 kg, respectively. There was difference of 13.60 kg between ILW and FLW means. The ILW and FLW were shown to positively correlate with body characteristics, and this correlation was statistically significant (P

Impacts of large scale climate drivers on precipitation in Sindh, Pakistan using machine learning techniques

Sindh province of Pakistan has a long history of severe droughts. Several large scale climate drivers (LSCD) are known for their effect on precipitation worldwide but studies in the Sindh region are missing; wide variety of LSCDs and lagged associative information. This study aimed to identify the significant LSCDs in Sindh province of Pakistan and improve the forecast skill of monthly precipitation by employing the principal component analysis (PCA), artificial neural network (ANN), Bayesian regularization neural network (BRNN), and multiple regression analysis (MRA), while considering the 12 months lagged LSCDs such as Nino-1+2, Nino-3, Nino-3.4, Nino-4, Quasi-Biennial Oscillation (QBO) at 30 and 50hPa (QBOI and QBOII), sea surface temperature (SST), 2m air temperature (T2M), 500 hPa and 850 hPa geopotential heights (H500 and H850), surface and 500 hPa zonal velocity (SU and U500), latent and sensible heat fluxes over land (LHFOL and SHFOL), and surface specific humidity (SSH). Global Land Data Assimilation System (GLDAS), Tropical Rainfall Measuring Mission (TRMM), ModernEra Retrospective Analysis for Research and Application (MERRA-2), NOAA, Freie University Berlin, and Hadley Centre Sea Ice and Sea Surface Temperature (HadISST) datasets were used. Results manifested that significant LSCDs with 99% confidence level were SU, U500, T2M, SST, SHFOL, LHFOL, SSH, and H850. During test period, compared with MR models of 0.39 to 0.64 and principal components of 0.31 to 0.57, the ANN and BRNN models had better predictive skills with correlation coefficients of 0.57 to 0.83 and 0.52 to 0.76, respectively. It can be concluded that the ANN and BRNN models enable us to predict monthly precipitation in Sindh region with lagged LSCDs.

Cyclic Direct Shear Testing of a Sand with Waste Tires

This study investigates the cyclic behavior of sand mixed with waste tires by using a series of strain-controlled cyclic direct shear tests under constant normal load (CNL) conditions. Crushed Stone Sand (CSS) was used in the experimental studies. The sand grains have angular shapes and sizes changing from 1.0 mm to 2.0 mm. Two different types of waste tires were used in the experiments; (i) tire crumb (TC), and (ii) tire buffing (TB). The TC grains have an angular shape and size between 1.0 mm and 2.0 mm, whereas TB grains used were found to be fiber-shaped, with dimensions changing from 1 mm to 9 mm, and an aspect ratio of about 1:5. The tests were carried out under 100 kPa vertical effective stress on the sand with 0%, 2.5%, 5%, 7.5%, and 10% waste tire contents. The testing results were found to be highly dependent on both the type and amount of waste tires in the mixtures. Furthermore, the behavior of the mixtures was estimated by the Bayesian Regularization Neural Network (BRNN) prediction model, for further use by researchers. The performance of the proposed BRNN model was found to provide a quite high correlation coefficient (R2 = 0.96).

A new avenue to improve the performance of integrated modeling for flash flood susceptibility assessment: Applying cluster algorithms

Flash flood is one of the most severe natural disasters around the world, and has caused sizeable economic losses and countless death. Assessing flash flood susceptibility by hybrid models of statistical and machine learning methods is essential for flood mitigation strategies and disaster preparedness. Although classifying the flash flood conditioning factors becomes a crucial step before applying these hybrid models, their impact on the accuracy of integrated modeling is still unclear. Most previous studies used natural break classification (NBC) and quantile classification methods by default to conduct the classification, but more classification methods have not been tried. In this context, this study introduced three clustering algorithms of K-Means, Expectation Maximization, and ISOMaximum likelihood algorithm (ISOMax) into the classification of factors, and compared them to NBC and quantile classification. To test the impact of classification methods on integrated modeling, these classification results were applied into the construction of three hybrid models (i.e., the integrating of frequency ratio with support vector machines, random forest, and bayesian-regularization neural networks). Then, the accuracy of these hybrid models was evaluated by using ROC curves and statistical indicators. The classification results show that the clustering intervals in the same factor varied with classification algorithms. It can be found from the model performance evaluation results that different classification algorithms will lead to discrepancies in accuracy of integrated modeling. Compared to NBC, the ISOMax allows a better fitting and prediction ability of hybrid models in this study. The application of clustering algorithm provides a new perspective for improving the accuracy of integrated modeling.

Digital Mapping of Soil Organic Carbon Based on Machine Learning and Regression Kriging.

In the last two decades, machine learning (ML) methods have been widely used in digital soil mapping (DSM), but the regression kriging (RK) model which combines the advantages of the ML and kriging methods has rarely been used in DSM. In addition, due to the limitation of a single-model structure, many ML methods have poor prediction accuracy in undulating terrain areas. In this study, we collected the SOC content of 115 soil samples in a hilly farming area with continuous undulating terrain. According to the theory of soil-forming factors in pedogenesis, we selected 10 topographic indices, 7 vegetation indices, and 2 soil indices as environmental covariates, and according to the law of geographical similarity, we used ML and RK methods to mine the relationship between SOC and environmental covariates to predict the SOC content. Four ensemble models-random forest (RF), Cubist, stochastic gradient boosting (SGB), and Bayesian regularized neural networks (BRNNs)-were used to fit the trend of SOC content, and the simple kriging (SK) method was used to interpolate the residuals of the ensemble models, and then the SOC and residual were superimposed to obtain the RK prediction result. Moreover, the 115 samples were divided into calibration and validation sets at a ratio of 80%, and the tenfold cross-validation method was used to fit the optimal parameters of the model. From the results of four ensemble models: RF performed best in the calibration set (R2c = 0.834) but poorly in the validation set (R2v = 0.362); Cubist had good accuracy and stability in both the calibration and validation sets (R2c = 0.693 and R2v = 0.445); SGB performed poorly (R2c = 0.430 and R2v = 0.336); and BRNN had the lowest accuracy (R2c = 0.323 and R2v = 0.282). The results showed that the R2 of the four RK models in the validation set were 0.718, 0.674, 0.724, and 0.625, respectively. Compared with the ensemble models without superimposed residuals, the prediction accuracy was improved by 0.356, 0.229, 0.388, and 0.343, respectively. In conclusion, Cubist has high prediction accuracy and generalization ability in areas with complex topography, and the RK model can make full use of trends and spatial structural factors that are not easy to mine by ML models, which can effectively improve the prediction accuracy. This provides a reference for soil survey and digital mapping in complex terrain areas.

A Stochastic Bayesian Neural Network for the Mosquito Dispersal Mathematical System

The objective of this study is to examine numerical evaluations of the mosquito dispersal mathematical system (MDMS) in a heterogeneous atmosphere through artificial intelligence (AI) techniques via Bayesian regularization neural networks (BSR-NNs). The MDMS is constructed with six classes, i.e., eggs, larvae, pupae, host, resting mosquito, and ovipositional site densities-based ODEs system. The computing BSR-NNs scheme is applied for three different performances using the data of training, testing and verification, which is divided as 75%, 15%, 10% with twelve hidden neurons. The result comparisons are provided to check the authenticity of the designed AI method portrayed by the BSR-NNs. The AI based BSR-NNs procedure is executed to reduce the mean square error (MSE) for the MDMS. The achieved performances are also presented to validate the efficiency of BSR-NNs scheme using the process of MSE, correlation, error histograms and regression.

Optimizing cellulase production from Aspergillus flavus using response surface methodology and machine learning models

The study aims to optimize cellulase (CMCase) production by Aspergillus flavus using wheat straw, an abundantly available lignocellulosic waste, as a substrate. Three parameters, i.e., nitrogen content (0.25 to 1%), fungal inoculum (0.25 to 1%), and duration (3 to 12 days), were optimized for maximum CMCase production using Response surface methodology-Box Behnken design (RSM-BBD). The quadratic response surface was suitable, and the model was significant. However, higher-order machine learning (ML) models were applied as the RSM-BBD model had a low R 2 value (0.85) and negative predicted R 2 value (−0.82). The supervised ML regression models, i.e., Artificial neural network (ANN) with Bayesian Regularization Neural Network (BRNN) and Radial Basis function Neural Network (RBFNN), Support vector machine (SVM) with Polynomial kernel (SPK), and Gaussian kernel (SGK), and Gaussian process learner (GPL) with the exponential kernel (GEK) and squared exponential kernel (GSEK) were applied. The RBFNN was the best performing model with a mean squared error (MSE) value of 0.0025 and an R 2 value of 0.98. The maximum CMCase production of 13.89 U/gds was at yeast extract 0.25%, fungal inoculum 0.625%, and duration of 12 days. There was almost a threefold increase in CMCase production after optimization compared to the screening experiments (4.7 U/gds). • Cellulase production by Aspergillus flavus was optimized. • Response surface and Machine learning models were used for optimization. • Machine learning models performed better than response surface methodology. • Artificial neural network-Radial basis function network model performed the best. • Optimization resulted in almost 3-fold increase in cellulase production.

Hybrid BRNN-ARIMA Model for Financial Time Series Forecasting

The accurate forecasting of time series is difficult and for exchange rate more difficult as well. Because it is difficult to predict as they continuously fluctuate during trading hours. Exchange rate forecasting plays a vital financial problem in recent era. It is extensively acknowledged that exchange rate stability implies that macroeconomic stability. In this study, a hybrid model is proposed to forecast exchange rates. Bayesian regularized neural network (BRNN) model is assembled with Autoregressive integrated moving average model (ARIMA) and develop hybrid BRNN-ARIMA model. Furthermore, the comparison of proposed hybrid model has been done with standalone BRNN, standalone ARIMA and random walk model. Quarterly exchange rate data from 1970Q1 to 2021Q2 of six countries comprises developed (UK, Canada, and Singapore) and developing (Pakistan, India, and Malaysia) are forecast. To evaluate the performance of these models RMSE, MAE and MAPE are applied. The results indicate that the proposed hybrid BRNN-ARIMA model outperforms the other studied model in forecasting exchange rates.

Using spectral indices and terrain attribute datasets and their combination in the prediction of cadmium content in agricultural soil

The continuous demand placed on farmland to yield optimal harvest is dependent on the continuous application of agrochemicals and fertilizers to increase soil fertility and manage diseases. Successive application of fertilizers and use of agrochemicals coupled with metal and steel industries introduce potentially toxic elements into the soil. Active agricultural activities and industrial emissions that result in atmospheric cadmium (Cd) injection and active deposition on agricultural soil (particularly from the primary metal industry, steel and iron industrial production). The concentration of cadmium in the study area exceeds the local background value. As a result, excessive cadmium soil concentration will contribute to increased toxic and carcinogenic effects, with negative implications for both environmental and human health. Therefore, determining the spatial distribution of Cd is critical for environmentally friendly agricultural production and reducing Cd emission into soils. The goals of this study are to (i) determine the variability of Cd prediction in agricultural soil using spectral indices or terrain attributes coupled with modeling algorithms, and (ii) determine whether combining spectral indices and terrain attributes coupled with modeling algorithms can improve Cd prediction efficiency in agricultural soil. The study applied three modelling scenarios, comprised prediction using terrain attributes coupled with digital soil mapping (DSM) approaches (Scenario 1), prediction using spectral indices combined with DSMs (Scenario 2), and prediction using a combination of terrain attributes, spectral indices, and DSMs (Scenario 3). Gaussian process regression (GPR), partial least square regression (PLSR), extreme gradient boosting (EGB), multivariate adaptive regression splines (MARS), Bayesian regularized neural network (BRNN), regularized random forest (RRF), Bayesian generalized linear model (BGLM), and M5 tree models were the DSMs used in the study. The M5 tree model and terrain attributes {Scenario 1 R2 = 0.77, concordance correlation coefficient (CCC) = 0.73, root mean square error (RMSE) = 0.45, mean absolute error (MAE) = 0.37 and median absolute error (MdAE) = 0.35}, EGB and spectral indices {Scenario 2, R2 = 0.83, CCC = 0.76, RMSE = 0.54, MAE = 0.33 and MdAE = 0.23} and the M5 tree model, spectral indices and terrain attributes {Scenario 3, R2 = 0.84, CCC = 0.81, RMSE = 0.39, MAE = 0.31 and MdAE = 0.24} were the overall best combinations that predicted Cd in the agricultural soil. The overall evaluation of the approaches suggested that the combination of spectral indices, terrain attributes, and the M5 tree model in Scenario 3 was the optimal technique for predicting Cd in agricultural soil. Thus, a combination of environmental covariates with a high correlation with the response variable, combined with appropriate modeling techniques predicting potentially toxic elements in agricultural soil, will produce the best results.

Evaluation of Machine-Learning Models for Predicting Aeolian Dust: A Case Study over the Southwestern USA

Aeolian dust has widespread consequences on health, the environment, and the hydrology over a region. This study investigated the performance of various machine-learning (ML) models including Multiple Linear Regression (MLR), Support Vector Machines (SVM), Random Forests (RF), Bayesian Regularized Neural Networks (BRNN), and Cubist (Cu) in predicting dust emissions over the Southwestern United States (US). Six meteorological and climatic variables (precipitation, air temperature, wind speed, ENSO, PDO, and NAO) were used to predict dust emissions. The correlation (r) and root mean square error (RMSE) for fine dust vary from 0.67 to 0.80, and 0.40 to 0.52 µg/m3, respectively. For coarse dust, the r and RMSE vary from 0.69 to 0.73, and 2.01 to 2.34 µg/m3, respectively. The non-linear ML models outperformed linear regression for both fine and coarse dust. ML models underestimated high concentrations of dust. Machine-learning models better predict fine dust than coarse dust over the Southwestern USA. Air temperature was found to be the most important predictor, followed by precipitation, for both fine- and coarse- dust-prediction over the region. These results improve our understanding of the predictability of Southwestern US dust.