Search

A robust short-term significant wave height (Hs) modelling framework based on an ensemble local mean decomposition method integrated with random forest (i.e. , En-RLMD-RF) is developed. The robust local mean decomposition (RLMD) decomposed the Hs data series into three subseries; amplitude modulation, frequency modulation and the low-frequency product function (PFs). The partial autocorrelation function was employed to determine the correlation-based significant predictor signals between the PFs at t 0 and t 1. Then the statistically significant PFs were incorporated into the random forest (RF) to construct the RLMD-RF model. The RLMD-RF based forecasted PFs were used again in the RF model as input predictors resulting in an ensemble-based RLMD-RF (i.e. , En-RLMD-RF) model for forecasting short-term Hs. The En-RLMD-RF model is validated and compared with RF, extreme learning machine (ELM) and multiple linear regression (MLR) models and their hybrids RLMD-RF, RLMD-ELM, RLMD-MLR, En-RLMD-ELM and En-RLMD-MLR counterparts using a set of performance metrics. The results demonstrated that the En-RLMD-RF model generates better forecasting accuracy against the benchmarking models. This study is beneficial for the application and optimization of more clean energy resources worldwide for sustained energy generation. [ABSTRACT FROM AUTHOR]

• A spatiotemporal risk prediction of wildlife-vehicle collisions (WVCs) is presented. • A Bavarian dataset of around 731,000 accidents over a period of 10 years (2010–2019) was used. • Heterogenous datasets of impact factors are pre-processed using GIS. • Different machine learning (ML) approaches are applied - random forest, feedforward neural networks, and support vector machine classifier to predict the risk of WVCs. • ML algorithms provide high risk prediction quality of over 86% (double-weighted sensitivity and single-weighted specificity rate). • Spatial transferability and temporal forecast of WVC prediction was verified. Introduction: The technology in the automotive industry is becoming increasingly safer in the age of automated driving, but the number of accidents is still high, especially in wildlife-vehicle collisions (WVCs). To better avoid these accidents, patterns involved in these accidents must be detected. Method : This paper presents a spatiotemporal risk prediction of WVCs, including various road and environmental characteristics. A process of data preparation using GIS automated by Python scripts was developed to enable a spatiotemporal link of diverse features for the subsequent predictive data analysis. Different machine learning (ML) approaches were applied- random forest (RF), feedforward neural networks (FNN), and support vector machine classifier (SVM) - including automated ML to predict the risk of WVCs. Therefore, a dataset of approximately 731,000 accidents reported to the police in Bavaria over a period of 10 years (2010–2019) was used. In addition, non-accidents were randomly generated in Python over time and space for the supervised ML processes. As the actual risk probability for WVCs and non-WVCs is not entirely known, the impact of different training ratios between accidents and non-accidents was tested on the risk prediction quality (RPQ) (25%, 50%, 75%, 90% WVCs) of the double-weighted sensitivity and single-weighted specificity rate. Results: The test yielded high mean values of RPQ as an indicator for a suitable WVC prediction. Both RF (86.6%) and FNN (86.7%) were identified as suitable choices for WVC risk prediction in terms of RPQ. The SVM yielded a lower prediction quality, even though acceptable results could be achieved within a shorter runtime. Practical Applications: Spatial transferability was verified since the algorithm was trained on the dataset of Bavaria (excluding Upper Bavaria) and successfully tested in Upper Bavaria. WVC forecasts were also proven through training with datasets from 2010-2017 and in prediction for 2018 and 2019. [ABSTRACT FROM AUTHOR]

• Performance assessment and production estimation of grid-connected PV station. • Strong relationship between weather variables and the output of the PV station. • Strong interdependence of the meteorological parameters and performance parameters. • The distinction between the different variable correlation results is a hard task. • Feature selection and PCA analysis were used as dimensional reduction techniques. • Dimensionality reduced data was used for training random forest models. Meteorological variables have an important effect on the performance of a grid-connected photovoltaic station, in this paper, the impact of meteorological variables on the 6 mWp grid-connected photovoltaic station in the desert of Adrar has been explored through performance assessment and output power forecasting. The impact of the meteorological variables on performance parameters has been investigated by performing interdependence and correlation studies. A clear interdependence between some variables has been observed, but the complete separation between each variable correlation effect has proven to be a difficult task. A combination of random forest method and pre-processing techniques namely feature selection and Principal component analysis has been developed in order to predict power production using meteorological variables as inputs. The forecasting models have been evaluated in terms of computation time, accuracy, and several statistical indicators. [ABSTRACT FROM AUTHOR]

The scale-up of nanocellulose production, a still challenging task, could benefit greatly from machine learning. In this paper, random forest regressors were applied to predict two important parameters for nanocellulose processing: nanofibrillation yield and aspect ratio. The dataset comprised diverse pulp sources, ranging from non-woody and woody feedstock, and two different types of pulp pre-treatments, namely mechanical refining and enzymatic hydrolysis. The obtained models were evaluated in terms of number of features and optimized with regard to some important hyperparameters, using the Bayesian optimization function from the SkOpt library. Considering the heterogeneity of the data, the prediction of both aspect ratio and yield of nanofibrillation was deemed satisfactory. For the former, R² scores of 0.986, 0.902 and 0.932 were obtained for the training, validation and testing subsets, respectively. Regarding the latter, R² scores of 0.994, 0.947 and 0.877 were achieved for the same types of subsets. Other important remarks are that enzyme dosage was the most important feature for predicting aspect ratio, while it did not play an important role for yield modelling. For this output, the transmittance at 600 nm accounted for almost 90% of the total feature importance. [Display omitted] • Aspect ratio and nanofibrillation yield were predicted by random forest regressors. • A 140-point dataset was gathered from experiments from different woods and crops. • Both independent variables and directly measurable responses constituted the inputs. • Bayesian optimization helped attain R2 of 0.90–0.95 for the validation subsets. • The kind of feedstock had lower feature importance than the type of pretreatment. [ABSTRACT FROM AUTHOR]

Background: Depression and post-traumatic stress disorder (PTSD) are among the most common psychiatric disorders observed in children and adolescents exposed to sexual abuse.Objective: The present study aimed to investigate the effects of many factors such as the characteristics of a child, abuse, and the abuser, family type of the child, and the role of social support in the development of psychiatric disorders using machine learning techniques.Participants and Settings: The records of 482 children and adolescents who were determined to have been sexually abused were examined to predict the development of depression and PTSD.Methods: Each child was evaluated by a child and adolescent psychiatrist in the psychiatric aspect according to the DSM-V. Through the data of both groups, a predictive model was established based on a random forest classifier.Results: The mean values and standard deviation of the 10-k cross-validated results were obtained as accuracy: 0.82% (+/- 0.19%), F1: 0.81% (+/- 0.19%), precision: 0.81% (+/- 0.19%), recall: 0.80% (+/- 0.19%) for children with depression; and accuracy: 0.72% (+/- 0.12%), F1: 0.71% (+/- 0.12%), precision: 0.72% (+/- 0.12%), recall: 0.71% (+/- 0.12%) for children with PTSD, respectively. ROC curves were drawn for both, and the AUC results were obtained as 0.88 for major depressive disorder and 0.76 for PTSD.Conclusions: Machine learning techniques are powerful methods that can be used to predict disorders that may develop after sexual abuse. The results should be supported by studies with larger samples, which are repeated and applied to other risk groups. [ABSTRACT FROM AUTHOR]

The stopping power of a material upon interaction with an energetic ion is the key measure of how far that ion will travel. The implications of accurate particle range calculations are tremendous, affecting every single application in which particle radiation is involved, from nuclear power to medicine. An approach is presented which attempts to overcome current shortcomings in the theoretical understanding of stopping power, as well as the methods used to interpret and exploit measured data. This is a considerable challenge, however the use of a novel machine learning methodology is shown to hold great promise in this endeavour: the ultimate aim being the ability to correctly predict the stopping value for any energy, ion and target combination, having no pre-existing experimental data. A random forest regression algorithm is trained using over 34,000 experimental measurements, representing stopping power values for 522 ion-target combinations across the energy range 10 - 3 to 10 2 MeV/amu, and ion and target atomic masses 1 to >240. Evaluation is carried out using several fundamental error metrics, over the whole dataset as well as for individual combinations, to provide the most comprehensive understanding of performance when tested under strict cross-validation criteria. The resulting model is shown to yield predicted stopping power curves corresponding closely to those of the true experimental values, with an ability to generalise across target elements, compounds, mixtures, alloys and polymers, irrespective of phase, and for a wide range of ion masses. [ABSTRACT FROM AUTHOR]

When a forecast of the total value over several time periods ahead is required, forecasters are presented with two temporal aggregation (TA) approaches to produce required forecasts: i) aggregated forecast (AF) or ii) aggregate data using non-overlapping temporal aggregation (AD). Often, the recommendation is to aggregate data to a frequency relevant to the decision the eventual forecast will support and then produce the forecast. However, this might not be always the best choice and we argue that both AF and AD approaches may outperform each other in different situations. Moreover, there is a lack of evidence on what indicators may determine the superiority of each approach. We design and execute an empirical experiment framework to first explore the performance of these approaches using monthly time series of M4 competition dataset. We further turn the problem into a classification supervised learning by constructing a database consisting of features of each time series as predictor and model class labelled as AF/AD as response/outcome. We then build machine learning algorithms to investigate the association between time series features and the performance of AF and AD. Our findings suggest that both AF and AD approaches may not consistently generate accurate results for every individual series. AF is shown to be significantly better than AD for the monthly M4 time series, especially for longer horizons. We build several machine learning approaches using a set of extracted time series features as input to predict accurately whether AD or AF should be used. We find out that Random Forest (RF) is the most accurate approach in correctly classifying the outcome assessed both by statistical measures such as misclassification error, F-statistics, area under the curve, and a utility measure. The RF approach reveals that curvature, nonlinearity, seas_pacf, unitroot_pp, mean, ARCHM.LM, Coefficient of Variation, stability, linearity, and max_level_shif are among the most important features in driving the predictions of the model. Our findings indicate that the strength of trend, ARCH.LM, hurst, autocorrelation lag 1, unitroot_pp, and seas_pacf may favour AF approach, while lumpiness, entropy, nonlinearity, curvature, and strength of seasonality may increase the chance of AD performing better. We conclude the study by summarising the findings and present an agenda for further research. [ABSTRACT FROM AUTHOR]

Highlights • Machine learning models are compared for simulation and forecasting of streamflows. • Support Vector Regression, Artificial Neural Networks and Random Forest are used. • A base flow separation method is coupled with machine learning models. • North Fork, Chehalis, Carson and Sacramento rivers in the United States are analysed. • Base flow separation method improved streamflow simulation to a certain degree. Abstract Efficient simulation of rainfall-runoff relationships is one of the most complex problems owing to the high number of interrelated hydrological processes. It is well-known that machine learning models could fail in simulating streamflows from only meteorological variables in the absence of antecedent streamflow values. The main reason for this could be low and lagged relationships between streamflow and meteorological variables. To overcome this inefficiency, for the first time, we developed a simulation framework by coupling a base flow separation method to three machine learning methods. It was demonstrated that separating streamflow into different components such as base flow and surface flow can be useful for improving simulation and forecasting capabilities of machine learning models. We simulated streamflow in four rivers in the United States with Support Vector Regression (SVR), Artificial Neural Networks (ANNs) and Random Forest (RF) as a function of precipitation (P), temperature (T) and potential evapotranspiration (PET). We concluded that the base flow separation method improved the simulation performances of the machine learning models to a certain degree. Apart from the simulation scheme, we also employed a forecasting scheme by using the antecedent observed discharge values in addition to P, T, and PET. We discussed performances of models in simulation and forecasting of streamflow regarding model types, input structures and catchment dynamics in detail. [ABSTRACT FROM AUTHOR]

Fuel consumption influences both the economic and environmental perspectives of shipping. With the help of machine learning, meaningful knowledge and complex relationships can be extracted from high-dimensional historical data. In this study, machine learning models were developed to predict the fuel consumption of harbour vessels with ship-related and meteorological factors. The superiority of machine learning models over statistical linear regression model (Ridge regression) has been proved. This study further investigated whether the use of meteorological factors enhances the prediction of fuel consumption. A case study on the prediction of tugboat fuel consumption was conducted. The Random Forest model outperformed the other models. Comparative experiments showed that meteorological factors collectively add value to the fuel consumption prediction, which enhances the accuracy from 0.7 to 38.9%. The potential uses of the prediction results are highlighted in terms of both operational management and environmental evaluation aspects. • Machine learning models were developed to predict the fuel consumption of harbour vessels. • The superiority of machine learning models over statistical linear regression model (Ridge regression) has been proved. • The Random Forest model outperformed the other models. • The meteorological factors collectively add value to fuel consumption prediction, which enhances the accuracy from 0.7 to 38.9%. [ABSTRACT FROM AUTHOR]

• Precipitation δ 18O predicted using geostatistical and machine learning algorithms. • Prediction performance evaluated using ordinary and hybrid error metrics. • Machine learning algorithms performed better, particularly using scarcer datasets. • Two spatiotemporally different subcontinental datasets of precipitation δ 18O were used. Stable isotopes of precipitation are important natural tracers in hydrology, ecology, and forensics. The spatially explicit predictions of oxygen and hydrogen isotopes in precipitation are obtained through different interpolation techniques. In the present study we aim to examine the performance of various interpolation techniques when predicting the spatial distribution of precipitation stable isotopes. The efficiency of combined geostatistical tools (i.e. regression kriging; RK) and various machine learning methods (including regression enhanced random forest methods: MRRF, RERF) are compared in interpolating the spatial variability of precipitation stable oxygen isotope values from two different sampling networks in Europe. To assess the performance of the models, mean squared error (MSE), nonparametric Kling Gupta efficiency (KGE), absolute differences and relative mean absolute error metrics were employed. It was found that the combination of the different regression techniques with Random Forest can produce estimations with comparable accuracy in terms of descending order of overall average MSE, MRRF: 2.61, RK: 2.77, RERF: 2.99, RF: 3.08. The best performing combined random forest model variant (MRRF) outperformed regression kriging in terms of a hybrid error metric (KGE) by 7.5%. Sequential random rarefying the station networks showed that machine-learning methods are more capable of maintaining high prediction accuracy even with fewer input data. This can be a great advantage when a suitable method is needed to predict the stable isotope composition of precipitation for large spatial domains where the spatial density of data stations shows large differences. [ABSTRACT FROM AUTHOR]

[Display omitted] • Machine learning methods were employed to predict bed expansion ratio. • XGBoost model has good regression and prediction ability with R2 higher than 0.97. • The relative importance of operating variables on bed expansion ratio was obtained. • Particle size has the strongest effect on bed expansion ratio. Liquid–solid fluidizations are mainly affected by particle density, particle size, particle shape, superficial velocity, liquid viscosity, etc. Owing to these complex high-dimensional impact processes, rapid and accurate prediction of fluidization characteristics is necessary but, still challenging. In this study, machine learning models were developed and employed to predict bed expansion ratio. Machine learning models, namely, linear, random forest, and XGBoost regression models, were trained using a measured 183- bed expansion ratio dataset under different operating conditions. By comparison, the XGBoost model had strongest regression and prediction ability with R2 higher than 0.97 and the predictions were very time-saving with calculation time less than 1 s. Finally, the bed expansion characteristics were analyzed and the operating conditions were evaluated by the developed model. The relative importance of operating conditions on bed expansion ratios was particle size > particle density > superficial velocity. [ABSTRACT FROM AUTHOR]

This paper analyzes the performance of three types of statistical models and a well-known physics-based model for forecasting the wave energy flux. The forecasts are run over horizons of 1–24 h at five buoys located in the Bay of Biscay. The data resolution is hourly. The 1999–2005 timeframe is used to train the models. The forecasts are run and evaluated over the six-year period from 2006 to 2012. The statistical forecasting models use three techniques: analogues, random forests (a machine learning algorithm) and a combination of the two. The physics model is the Wave Model (WAM). The forecasts are compared at a 95% confidence level with the simplest prediction – Persistence – and also with the nearest grid point of the WAM forecasts. Over horizons between 3 and 16–19 h at locations near the coast (where wave farms may be installed), the random forests models outperform the others, including WAM and Persistence. These models exploit the inherent predictability associated with the strong autocorrelation present in ocean energy values. The additional prognostic capabilities that random forests models provide over Persistence, are due to their ability to sucessfully incorporate the information that both, atmospheric and sea-state variables provide. [ABSTRACT FROM AUTHOR]

[Display omitted] • We built a random forest model on ozone using TROPOMI and HRRR data. • Our model had a cross-validated (CV) R 2 of 0.84 for daily ground ozone concentrations. • Model predictions captured daily ozone distributions at 10 km spatial resolution. • TROPOMI column O 3 data improved the characterization of high concentrations. Estimating ground-level ozone concentrations is crucial to study the adverse health effects of ozone exposure and better understand the impacts of ground-level ozone on biodiversity and vegetation. However, few studies have attempted to use satellite retrieved ozone as an indicator given their low sensitivity in the boundary layer. Using the Troposphere Monitoring Instrument (TROPOMI)'s total ozone column together with the ozone profile information retrieved by the Ozone Monitoring Instrument (OMI), as TROPOMI ozone profile product has not been released, we developed a machine learning model to estimate daily maximum 8-hour average ground-level ozone concentration at 10 km spatial resolution in California. In addition to satellite parameters, we included meteorological fields from the High-Resolution Rapid Refresh (HRRR) system at 3 km resolution and land-use information as predictors. Our model achieved an overall 10-fold cross-validation (CV) R 2 of 0.84 with root mean square error (RMSE) of 0.0059 ppm, indicating a good agreement between model predictions and observations. Model predictions showed that the suburb of Los Angeles Metropolitan area had the highest ozone levels, while the Bay Area and the Pacific coast had the lowest. High ozone levels are also seen in Southern California and along the east side of the Central Valley. TROPOMI data improved the estimate of extreme values when compared to a similar model without it. Our study demonstrates the feasibility and value of using TROPOMI data in the spatiotemporal characterization of ground-level ozone concentration. [ABSTRACT FROM AUTHOR]

People typically spend most of their time indoors. It is of importance to establish prediction models to estimate PM 2.5 concentration in indoor environments (e.g., residential households) to allow accurate assessments of exposure in epidemiological studies. This study aimed to develop models to predict PM 2.5 concentration in residential households. PM 2.5 concentration and related parameters (e.g., basic information about the households and ventilation settings) were collected in 116 households during the winter and summer seasons in Hong Kong. Outdoor PM 2.5 concentration at households was estimated using a land-use regression model. The random forest machine learning algorithm was then applied to develop indoor PM 2.5 prediction models. The results show that the random forest model achieved a promising predictive accuracy, with R2 and cross-validation R2 values of 0.93 and 0.65, respectively. Outdoor PM 2.5 concentration was the most important predictor variable, followed in descending order by the household marked number, outdoor temperature, outdoor relative humidity, average household area and air conditioning. The external validation result using an independent dataset confirmed the potential application of the random forest model, with an R2 value of 0.47. Overall, this study shows the value of a combined land-use regression and machine learning approach in establishing indoor PM 2.5 prediction models that provide a relatively accurate assessment of exposure for use in epidemiological studies. Image 1 • A practical indoor PM 2.5 prediction framework was proposed. • The combined land-use regression and machine learning approach was applied. • External validation was proposed to evaluate the prediction model. • The random forest model outperformed the linear mixed-effect regression model. • The outdoor PM 2.5 concentration is the most important predictor variable. [ABSTRACT FROM AUTHOR]

• 1975 samples of varied soils were analyzed for fertility and elemental contents. • pXRF data was successfully used to rapidly predict soil fertility properties. • Random Forest outperformed other machine learning algorithms. • Prediction models allowed spatial distribution of soil fertility properties. • Inexpensive and environment-friendly soil fertility prediction was achieved by pXRF. Conventional soil fertility analyses are laborious, costly, time-consuming, and produce hazardous waste. The high demand of these laboratory-based analyses prompted us to investigate an environment-friendly, rapid, and inexpensive methodology for soil fertility assessment. Portable X-ray fluorescence (pXRF) spectrometry allows the determination of total elemental concentration in soils quickly, simply and without hazardous waste production. However, incipient usage of this technology for the prediction of soil fertility properties has been reported for tropical conditions. Soil samples were collected from seven Brazilian states (n = 1975) aiming to use pXRF data to predict contents of available or exchangeable Ca2+, Mg2+, Al3+, K+ and P by testing different algorithms using 70% of the samples for model training, and the remaining 30% for model validation. In addition to point data predictions, the best performing models were applied to data obtained from a farm within the studied regions with a known cropping history to create soil fertility maps and illustrate another applicability of this approach. The attested use of pXRF data and machine learning algorithms stepwise Generalized Linear Model (GLM) and Random Forest (RF) to predict the contents of relevant soil fertility properties exhibited great potential. Validation of the models corroborated that RF resulted in more accurate predictions than GLM. Validation R2 values ranged from 0.59 to 0.82. Maps created were coherent with expected distributions of soil fertility attributes. This environment-friendly methodology may be used for the assessment of soil fertility properties in a wide range of tropical and subtropical soils with minimum waste generation and reduced costs. [ABSTRACT FROM AUTHOR]

The straightforward prediction for the air-entry value of compacted soils is practically useful, but the investigation on this issue is scarce. This study presents three alternative straightforward prediction models for the air-entry value of compacted soils using the representative machine learning algorithms of multi expression programming (MEP), evolutionary polynomial regression (EPR) and random forest (RF). Five known soil properties (i.e. sand content, fines content, plasticity index, initial water content and initial void ratio) are used as input variables. All models are developed based on a large database, covering a wide range of soil classifications. The results show that all the three proposed models are appropriate to predict the air-entry values of different compacted soils, with high prediction accuracies for both the training and the testing data. The monotonicity, the sensitivity and the robustness of the three prediction models are evaluated, showing consistency among different models with a slight difference and providing a strong support for the model feasibility. On the whole, the MEP and the EPR models are recommended for more convenient applications with explicit expression, while higher prediction accuracy may require the RF model although no explicit expression can be derived. • Air-entry value of compacted soils is precited using three machine learning algorithms (MEP, EPR, RF) with high precision. • The correctness and the feasibility of all the three models are verified. • The MEP and the EPR models are recommended for more convenient applications with high prediction precision. [ABSTRACT FROM AUTHOR]

Evaluating the efficacy of air pollution control policies is an essential part of the decision-making process to develop new policies and improve existing measures. Since 2005, Fujian Province of Mainland China and Taiwan across the Taiwan Strait have both implemented aggressive air pollution control policies designed based on different principles, but a comprehensive evaluation of these control policies on PM 2.5 pollution levels is still lacking. In the current study, we assessed the effects of these policies in the Taiwan Strait Region from 2005 to 2018 using full-coverage, high-resolution PM 2.5 generated by a satellite-driven machine learning model. A ten-fold cross-validation for our prediction model showed an R 2 value of 0.89, demonstrating that these predictions can be used for policy evaluation. During the 14-year period, PM 2.5 levels in all areas of Fujian and Taiwan underwent a significant decrease. Separate regression models for policy evaluation in Taiwan and Fujian showed that all considered policies have mitigated PM 2.5 pollution to various degrees. The Clean Air Action Plans (CAAP) is the most effective control policy in Taiwan, while the Action Plan of Air Pollution Prevention and Control (APPC-AP) and Three-year Action Plan for Blue Skies (3YAP-BS) as well as their provincial implementation plans are the most successful in Fujian. The effectiveness of control policies, however, varies by land-use types especially for Taiwan. • PM 2.5 levels were estimated in the Taiwan Strait Region at 1-km resolution from 2005 to 2018 • The most effective control policies to lower PM 2.5 levels in Taiwan and Fujian were analyzed • The effectiveness of control policies varies by land use type in Taiwan [ABSTRACT FROM AUTHOR]

The present study aims to compare the performance of two machine learning techniques that can unveil the relationship between the input and target variables and predict the ground vibration (peak particle velocity, PPV) due to quarry blasting. To this end, a Random Forest (RF) model and a Bayesian Network (BN) model were developed. Before developing these models, and in order to illustrate the necessity of proposing new intelligent systems, a new empirical equation is proposed, using maximum charge per delay and distance from the blast-face. The results confirm that there is indeed a need to develop intelligent systems with more input parameters. Thus, a Feature Selection (FS) model is applied to decrease the dimensionality of data and remove the irrelevant data. The outputs of this technique set five parameters, hole depth, power factor, stemming, maximum charge per delay and distance from the blast-face, as the most important model inputs necessary to predict PPV. After constructing FS-BN and FS-RF models and comparing them under different conditions (i.e., computational cost, accuracy and robustness), it is found that the developed FS-RF model can be introduced as a new model in the field of blasting environmental issues. The accuracy level of the FS-RF model is quite high; 92.95% and 90.32% for the train and test stages, respectively, while 92.95% and 87.09% accuracy is calculated for train and test of the FS-BN model. Thus, both developed hybrid models can effectively unveil the relationships between the input and target variables. • We measured the most effective parameters on peak particle velocity induced by blasting. • A new peak particle velocity empirical equation was proposed and compared with previous equations. • A feature selection model was applied to remove the irrelevant predictors. • Two hybrid models were developed to predict peak particle velocity. • A comparison between these models was made and the best among them was chosen. [ABSTRACT FROM AUTHOR]

• The study provides inspiration to use simulation data for yield predictive modeling. • The proposed machine learning algorithm approach has achieved 93.8% accuracy. • Interactions between bee species composition and weather are significant in yield variability. • Prediction showed that wet rainy springs will greatly reduce blueberry yield in the future. The most challenging task in the agricultural sector is to accurately predict crop yield. A typical machine learning algorithm often uses real data to predict crop yield. In this study, we used data generated by the Wild Blueberry Pollination Model, a spatially explicit simulation model validated by field observation and experimental data collected in Maine USA during the last 30 years. The main aim of this study is to evaluate the relative importance of bee species composition and weather factors in regulating wild blueberry agroecosystems. Specifically, we sought to reveal how bee species composition and weather affect yield and to predict optimal bee species composition and weather conditions that achieve the best yield using computer simulation and machine learning algorithms. Multiple linear regression (MLR), boosted decision trees (BDT), random forest (RF), and extreme gradient boosting (XGBoost) were evaluated as predictive tools. We also performed a predictor selection before submitting our data to the learning algorithms. In this way, we are able to reduce the dimension of the input without a significant drop in prediction accuracy. As a result, clone size, honeybee, bumblebee, Andrena bee species, Osmia bee species, maximum of upper-temperature ranges, and the number of days with precipitation were chosen as the best predictor variable subset. The results showed that the XGBoost outperformed other algorithms in all measures of model performance for predicting the yield of wild blueberry by achieving a coefficient of determination (R2) of 0.938, root mean square error (RMSE) of 343.026, mean absolute error (MAE) of 206 and relative root mean square error (RRMSE) of 5.444%. The results are consistent with previous work on predicting wild blueberry fruit yield using digital color photography by (Zaman et al., 2008). This study showed that crop yield predictions can be based on computer simulation modeling datasets. Therefore, if a reasonable prediction can be reached, this study should have a significant impact, especially when data collection in the field is challenging. [ABSTRACT FROM AUTHOR]