Model Performance In Terms Research Articles

Modeling regional-scale groundwater arsenic hazard in the transboundary Ganges River Delta, India and Bangladesh: Infusing physically-based model with machine learning

For the last few decades, toxic levels of arsenic (As) in groundwater from the aquifers of the Ganges River delta, India and Bangladesh, have been known to cause serious public health concerns. Innumerable studies have advocated the control of geomorphologic, geologic, hydrogeologic, biogeochemical, and anthropogenic factors on arsenic mobilization, flow, and distribution patterns within the Ganges River delta. We have developed transboundary regional-scale models for computing the probability of groundwater As concentrations to exceed the WHO permissible thresholds for drinking water of 10 μg/L within the Ganges River delta as a function of the various geomorphologic-(hydro)geologic-hydrostratigraphic-anthropogenic controlling factors, using statistical methods and artificial intelligence (AI) [i.e., machine learning] techniques namely, Random Forest (RF), Boosted Regression Trees (BRT) and Logistic Regression (LR) algorithms, followed by probabilistic delineation the high As-hazard zones within the delta. A “hybrid multi-modeling approach” was adapted for this study, which involved the introduction of hydrostratigraphic parameters (aquifer connectivity and surficial aquitard thickness) derived from a high-resolution transboundary hydrostratigraphic model developed for the Ganges River delta aquifer system, as predictors for modeling groundwater As probabilities within the delta.The RF model outperforms the BRT and LR model in terms of model performance. Model outputs suggest the dominant influence of surficial aquitard thickness and groundwater-fed irrigated area (%) on groundwater As. While, the north-central and southern regions of the Ganges River delta show low As-hazard (<10 μg/L), the western and north-eastern regions demonstrate elevated hazard level (>10 μg/L). An estimated 30.3 million people are found to be exposed to elevated groundwater As within the study area. Thus, our study demonstrates that such hybrid, predictive models are not only helpful in delineating the regional-scale distribution of groundwater As-hazard zones in the areas with limited As data but is also useful in identifying the possible exogenous forcing that may have led to the worst, natural pollution in human history.

Undergraduate Engineering Students’ Types and Quality of Knowledge Used in Synthetic Modeling

Modeling is an important element of discovery and design processes because it can help individuals to comprehend and facilitate solutions to problems, mediate among mental and external representations, and off-load cognitive demands. However, engaging in model generation, comprehension, and transformation requires the orchestration of domain knowledge, meta-representational cknowledge, and various reasoning processes. This study aims to understand the interplay between domain knowledge, meta-representational knowledge, and students’ reasoning processes while engaging in modeling activity. Specifically, our goal is to investigate: (a) the types and quality of knowledge students used when performing modeling activities, and (b) the differences in types of knowledge used by students regarding their performance in a modeling and simulation challenge. Our qualitative analysis focused on verbal descriptions via a retrospective semi-structured interview and artifacts created by seventeen students after engaging in synthetic modeling activity. The resulting analysis reports overall quality for the types of domain and meta-representational knowledge that students used along with reasoning processes during synthetic modeling activity. The analysis also compared and contrasted the patterns of conceptual understanding and problem-solving skills between three groups, which differed in terms of modeling performance.

Extreme Gradient Boosting Model for Rain Retrieval using Radar Reflectivity from Various Elevation Angles

The purpose of this study was to develop an optimal estimation model for rainfall rate retrievals using radar reflectivity, thereby gaining an effective grasp of rainfall information for disaster prevention uses. A process was designed for evaluating the optimal retrieval models using various dataset combinations with radar reflectivity and ground meteorological attributes. Various ground meteorological attributes (such as relative humidity, wind speed, precipitation, etc.) were obtained using the land-based weather stations affiliated with Taiwan’s Central Weather Bureau (CWB). This study used nine radar reflectivity provided by the Hualien weather surveillance radar station’s Volume Cover Pattern 21 system. The developed models are built using multiple machine learning algorithms, including linear regression (REG), support vector regression (SVR), and extreme gradient boosting (XGBoost), in addition to the Marshall–Palmer formula (MP). The study examined 14 typhoons that occurred from 2008 to 2017 at Chenggong station in southeast Taiwan, and Lanyu station in the outlying islands, and the top four major rainfall events were designated as test typhoons—Nanmadol (2011), Tembin (2012), Matmo (2014), and Nepartak (2016). The results indicated that for rainfall retrievals, radar reflectivity at a scanning (elevation) angle of 6.0° combined with ground meteorological attributes were the optimal input variables for the Chenggong station, whereas radar reflectivity at an elevation angle of 4.3° combined with ground meteorological attributes were optimal for the Lanyu station. In terms of model performance, XGBoost models had the lowest error index at Chenggong and Lanyu stations compared with MP, REG, and SVR models. XGBoost models at Lanyu station had the highest efficiency coefficient (0.903), and those at Chenggong station had the second highest (0.885). As a result, pairing the combination of optimal radar reflectivity and ground meteorological attributes, as verified by the evaluation process, with a high-efficiency algorithm (XGBoost) can effectively increase the accuracy of rainfall retrieval during typhoons.

Prediction of solar radiation on the horizon using neural network methods, ANFIS and RSM (case study: Sarpol-e-Zahab Township, Iran)

Solar energy is one of the clean and healthy energies. Due to the high cost of required equipment to convert solar energy into the desired form, the economic facets must be addressed and the equipment should be installed in areas with higher accessible solar energy. However, due to the complex and time-consuming process of calculating solar radiation, it seems necessary to develop more simple models with higher estimation capability. Therefore, the present study investigated the prediction of solar radiation on the horizon using neural network methods, ANFIS and RSM, in Sarpol-e-Zahab Township, Kermanshah, Iran. In this respect, the meteorological data of this township were collected. Then, the key parameters were selected by performing sensitivity analysis, and models were designed and optimized using ANFIS, ANN, and RSM methods. Moreover, respective correlation coefficients and mean square errors of each method were obtained (ANFIS (0.993 and 0.0005), ANN (0.996 and 0.00029), and RSM (0.996 and 0.00027), respectively). Also, the neural network and response surface methodology were superior to the ANFIS Model in terms of performance, simplicity, and speed. In short, the performance of the response surface methodology was slightly better than that of the neural network.

Artificial intelligence neural network based on intelligent diagnosis

The medical model that uses artificial intelligence technology to assist diagnosis and treatment is called smart medicine. It can learn the medical knowledge of experts, and can simulate the thinking and reasoning of doctors to give patients a reliable diagnosis and treatment plan. The purpose of this article is to help young doctors with uneven distribution of medical resources and insufficient experience by exploring the application of artificial intelligence neural networks in the intelligent diagnosis of diabetes. This paper proposes to use BP neural network and probabilistic neural network to model diabetes diagnosis. First, the number of hidden layer units of BP network is selected according to the input feature vector. The training effect is best when the number of hidden layer units is 12, and the diagnostic accuracy rate is 91.7% through experiments. Then built a PNN network model, using 75% of the data for training, 25% of the data for testing, and testing the effectiveness of the network model. The diagnostic accuracy was 97.9%. Finally, the accuracy of the 20 neural network models was tested 20 times. After a comparative analysis, it is concluded that the PNN network model is better than the BP neural network model in terms of performance and accuracy. Compared with the traditional diagnosis process, the neural network-based diagnosis model can effectively save doctors time and improve diagnosis efficiency.

Identifying Predictors of Suicide in Severe Mental Illness: A Feasibility Study of a Clinical Prediction Rule (Oxford Mental Illness and Suicide Tool or OxMIS).

BackgroundOxford Mental Illness and Suicide tool (OxMIS) is a brief, scalable, freely available, structured risk assessment tool to assess suicide risk in patients with severe mental illness (schizophrenia-spectrum disorders or bipolar disorder). OxMIS requires further external validation, but a lack of large-scale cohorts with relevant variables makes this challenging. Electronic health records provide possible data sources for external validation of risk prediction tools. However, they contain large amounts of information within free-text that is not readily extractable. In this study, we examined the feasibility of identifying suicide predictors needed to validate OxMIS in routinely collected electronic health records.MethodsIn study 1, we manually reviewed electronic health records of 57 patients with severe mental illness to calculate OxMIS risk scores. In study 2, we examined the feasibility of using natural language processing to scale up this process. We used anonymized free-text documents from the Clinical Record Interactive Search database to train a named entity recognition model, a machine learning technique which recognizes concepts in free-text. The model identified eight concepts relevant for suicide risk assessment: medication (antidepressant/antipsychotic treatment), violence, education, self-harm, benefits receipt, drug/alcohol use disorder, suicide, and psychiatric admission. We assessed model performance in terms of precision (similar to positive predictive value), recall (similar to sensitivity) and F1 statistic (an overall performance measure).ResultsIn study 1, we estimated suicide risk for all patients using the OxMIS calculator, giving a range of 12 month risk estimates from 0.1-3.4%. For 13 out of 17 predictors, there was no missing information in electronic health records. For the remaining 4 predictors missingness ranged from 7-26%; to account for these missing variables, it was possible for OxMIS to estimate suicide risk using a range of scores. In study 2, the named entity recognition model had an overall precision of 0.77, recall of 0.90 and F1 score of 0.83. The concept with the best precision and recall was medication (precision 0.84, recall 0.96) and the weakest were suicide (precision 0.37), and drug/alcohol use disorder (recall 0.61).ConclusionsIt is feasible to estimate suicide risk with the OxMIS tool using predictors identified in routine clinical records. Predictors could be extracted using natural language processing. However, electronic health records differ from other data sources, particularly for family history variables, which creates methodological challenges.

Heteroskedastic geographically weighted regression model for functional data

A large number of approaches for modelling spatially dependent functional variables often assume that the functional regression coefficients are constant over the region of interest. However, in many occasions it is far more realistic that functional coefficients vary at a local level. The present paper proposes a calibrated heteroskedastic geographically weighted regression model (H-GWR) in the functional framework. Our model assumes that the variance varies across the space, and that each local model (defined at each location) gives a local estimation of the variance. Since this assumption depends on the chosen distance between the focal point and the rest of spatial observations, we use a back-fitting approach to calibrate the H-GWR model with a parameter-specific distance metric. This new approach improves the model performance in terms of predictive fit, as illustrated by simulations and through the analysis of a financial real data set.

Numerical Simulation of Large Wave Heights from Super Typhoon Nepartak (2016) in the Eastern Waters of Taiwan

Super Typhoon Nepartak (2016) was used for this case study because it is the most intense typhoon that made landfall in Taiwan in the past decade. Winds extracted from the Climate Forecast System version 2 (CFSV2) and ERA5 datasets and merged with a parametric typhoon model using two hybrid techniques served as the meteorological conditions for driving a coupled wave-circulation model. The computed significant wave heights were compared with the observations recorded at three wave buoys in the eastern waters of Taiwan. Model performance in terms of significant wave height was also investigated by employing the CFSV2 winds under varying spatial and temporal resolutions. The results of the numerical experiments reveal that the simulated storm wave heights tended to decrease significantly due to the lower spatial resolution of the hourly winds from the CFSV2 dataset; however, the variations in the storm wave height simulations were less sensitive to the temporal resolution of the wind field. Introducing the combination of the CFSV2 and the parametric typhoon winds greatly improved the storm wave simulations, and similar phenomena can be found in the exploitation of the ERA5 dataset blended into the parametric wind field. The overall performance of the hybrid winds derived from ERA5 was better than that from the CFSV2, especially in the outer region of Super Typhoon Nepartak (2016).

Deep Reinforcement Learning for Multiobjective Optimization.

This article proposes an end-to-end framework for solving multiobjective optimization problems (MOPs) using deep reinforcement learning (DRL), that we call DRL-based multiobjective optimization algorithm (DRL-MOA). The idea of decomposition is adopted to decompose the MOP into a set of scalar optimization subproblems. Then, each subproblem is modeled as a neural network. Model parameters of all the subproblems are optimized collaboratively according to a neighborhood-based parameter-transfer strategy and the DRL training algorithm. Pareto-optimal solutions can be directly obtained through the trained neural-network models. Specifically, the multiobjective traveling salesman problem (MOTSP) is solved in this article using the DRL-MOA method by modeling the subproblem as a Pointer Network. Extensive experiments have been conducted to study the DRL-MOA and various benchmark methods are compared with it. It is found that once the trained model is available, it can scale to newly encountered problems with no need for retraining the model. The solutions can be directly obtained by a simple forward calculation of the neural network; thereby, no iteration is required and the MOP can be always solved in a reasonable time. The proposed method provides a new way of solving the MOP by means of DRL. It has shown a set of new characteristics, for example, strong generalization ability and fast solving speed in comparison with the existing methods for multiobjective optimizations. The experimental results show the effectiveness and competitiveness of the proposed method in terms of model performance and running time.

P070 Machine learning approaches to identify IBD biomarkers from longitudinal microbiome data

Abstract Background Inflammatory bowel disease (IBD) has been shown to be associated with alterations in the intestinal microbiome. However, the precise nature of these microbial changes remains unclear. With billions of microbes present within the gut, novel and powerful computational techniques are required to identify the relevant shifts in microbiota contributing to the disease. Machine learning (ML) allows a data-driven approach to identify these discrete dynamic changes, while the findings of the ML algorithms can be interpreted using systems biology (SB) techniques. By combining ML and SB approaches, we aim to characterise key microbial factors in IBD pathogenesis, distinct patterns of variability in a diverse patient cohort and provide a method for patient stratification. Methods The causal relationship between the changes in the gut microbiome and IBD is difficult to establish. Data from cross-sectional studies are plagued by confounding factors and inconsistencies between cohorts. To overcome this, the authors used rich longitudinal datasets and integrated metagenomic, multi-omic and clinical patient data. This workflow has been validated using large longitudinal IBD databases, including data from IBDMDB. We assessed the performance of the ML models using well-documented performance metrics to ensure the outcomes were robust. Results As a baseline, we used multiple ML models to predict disease type (UC, CD and non-IBD) from integrated multi-omics profiles. We analysed multiple ML techniques, including linear (e.g. linear mixed model), non-linear (e.g. Random Forest), time-series models (e.g. Rotation Forest) and deep learning models (e.g. long short-term memory network model). The authors identified the models which would allow flexibility to analyse the dynamic nature of the microbiome and allow integration of the microbiome data with clinical patient data. The payoff of greater flexibility was a reduction in the model performance in terms of identifying specific features from the metagenomics that could be used as biomarkers. However, we were able to identify connections between microbial and host proteins relevant to IBD and were able to stratify these by the patient’s metagenomic data. Conclusion We have developed an integrated ml-based microbiome analysis pipeline to identify biomarkers for IBD from longitudinal metagenomic data. Furthermore, using a variety of SB approaches, we were able to interpret the predicted key microbial features and communities by inferring connections between microbial and host proteins. This pipeline will enable us to analyse vast amounts of patient microbiome data in the context of clinical and metagenomic data, to allow identification of biomarkers for disease subtypes.

Optimal Portfolio Selection of Wind Power Plants Using a Stochastic Risk-Averse Optimization Model, Considering the Wind Complementarity of the Sites and a Budget Constraint

This work focuses on the best financial resources allocation to define a wind power plant portfolio, considering a set of feasible sites. To accomplish the problem formulation and solution, the first step was to establish a long-term wind series reconstruction methodology for generating scenarios of wind energy, applying it to study five different locations of the Brazilian territory. Secondly, a risk-averse stochastic optimization model was implemented and used to define the optimal wind power plant selection that maximizes the portfolio financial results, considering an investment budget constraint. In a sequence, a case study was developed to illustrate a practical situation of applying the methodology to the portfolio selection problem, considering five wind power plants options. The case study was supported by the proposed optimization model, using the scenarios of generation created by the reconstruction methodology. The obtained results show the model performance in terms of defining the best financial resources allocation considering the effect of the complementarity between sites, making it feasible to select the optimal set of wind power plants, characterizing a wind plant optimal portfolio that takes into account the budget constraint. The adopted methodology makes it possible to realize that the diversification of the portfolio depends on the investor risk aversion. Although applied to the Brazilian case, this model can be customized to solve a similar problem worldwide.

Evaluating the effects of soil erosion and productivity decline on soil carbon dynamics using a model-based approach

Abstract. Sustained accelerated soil erosion alters key soil properties such as nutrient availability, water holding capacity, soil depth and texture, which in turn have detrimental effects on crop productivity and therefore reduce C input to soils. In this study, we applied a 1-D soil profile model that links soil organic carbon (SOC) turnover, soil erosion and biomass production. We used observational data to constrain the relationship between soil erosion and crop productivity. Assuming no change in effort, we evaluated the model performance in terms of SOC stock evolution using published observational data from 10 catchments across Europe and the USA. Model simulations showed that accounting for erosion-induced productivity decline (i) increased SOC losses by 37 % on average compared to a scenario where these effects were excluded, and (ii) improved the prediction of SOC losses when substantial soil truncation takes place. Furthermore, erosion-induced productivity decline reduced soil–atmosphere C exchanges by up to 30 % after 200 years of transient simulation. The results are thus relevant for longer-term assessments and they stress the need for integrated soil–plant models that operate at the landscape scale to better constrain the overall SOC budget.

A Hyper Heuristics Technique for Data Partitioning and Scheduling to Heterogeneous Systems using Genetic Algorithm and Improved Particle Swarm Optimization

Development of the load partitioning for multiple round load distribution and effective scheduling of partitioned load to heterogeneous processor is primary goal of distributed and parallel system. In this paper, we propose hyper heuristics scheduling algorithm for load partitioning using genetic and improved particle swarm optimization techniques.A communication model is used to predict the optimal activation order, optimal number of processor and optimal number of rounds of the load. Heuristics Based Scheduling Algorithm is proposed using Hyper Heuristic Scheduling which is used to find the candidate solution (low level heuristic) to form Scheduling Solutions (heuristics algorithms) for large scale system with diversity operator as sequence dependent and sequence independent scheduling. For this solution, processing time of the entire processing load will be reduced. Hybrid Real Code genetic algorithm(HRGA) computes optimal activation order with cross over and mutation operator without considering the processor latency and different types of variation in the perturbation parameters. In order to optimize this issue, we utilizeImproved Particle swarm optimization (IPSO) determine the load fraction for generating activation order in terms of dynamically predicting fitness value of the processor with certain number. The Simulation analysis demonstrates the proposed model performance in terms of mean, standard deviation, computational complexity and Average Execution Time comparing against hybrid real coded genetic algorithm

Understanding the effects of soil data quality on SWAT model performance and hydrological processes in Tamedroust watershed (Morocco)

Hydrological modeling remains an indispensable tool for water resources research in order to understand the watershed functioning and to effectively manage water supplies. The study of the hydrologic functioning of a semi-arid watershed is a real challenge in the lack of reliable and regular data. In this work, we propose (1) to evaluate the performance of Soil and Water Assessment Tool (SWAT) on the TAMEDROUST watershed (Morocco) under extremely contrasted climatic conditions during the period 1998–2002 and (2) to test the effects of the soil data resolution on the watershed response. For this second purpose, two soil databases are compared: (i) HWSD-2L, a low-resolution soil database with three soil types obtained from the Harmonized World Soil Database produced by FAO and (ii) TAMED-SOL, a refined database with eleven soil types, performed from field measurements and laboratory analyses.Results before calibration showed a considerable variability and a significant effect of the soil characteristics on the different components of the hydrological cycle such as SW (soil water content) and GWQ groundwater discharge into reach). Flows obtained by HWSD-2L soil database were consistently lower than those simulated by TAMED-SOIL but both databases overpredicted streamflow if compared to observed data. To harmonize the data of the two databases, we proceeded to a harmonization of the soil depth by considering in HWSD-2L only the surface layer, thus creating a third database HWSD-1L which was also considered in the evaluation of the model response. After the calibration period, both soils databases (TAMED-SOIL and HWSD-2L) improved the model performance in terms of streamflow, with values of R2 and NSE between 0.64 and 0.65. Model validation was acceptable and similar for both databases with R2 and NSE values of 0.76 and 0.57.We can conclude that the quality and resolution of soil data affects all hydrological cycle components, mainly SW and water yield (WYLD), but do not lead to significant discrepancies in streamflow simulation, especially after the model calibration.

Identifying the factors associated with cesarean section modeled with categorical correlation coefficients in partial least squares.

Cesarean section (CS) is associated with maternal morbidity and mortality in developing countries. This study is conducted to assess factors associated with CS in Pakistan using partial least squares (PLS) algorithm, where categorical factors are modeled. Nationally representative maternal data from Pakistan Demographic and Health Surveys (PDHS) conducted during 2012-2013 is used in this study. Among correlation coefficient based PLS regression proposed algorithms for categorical factors, Pearson’s Contingency Coefficient (CC) PLS coupled with loading weight (LW) appeared to be the most efficient method in terms of model performance and influential factor selection. Region of residence, type of place of residence, mother’s and her partner’s level of education, wealth index, year of birth, previous terminated pregnancy, use of contraception, prenatal care provided by a doctor and nurse/midwife/LHV (lady health visitor), assistance provided by a nurse/midwife/LHV,number of antenatal visits, size of child, antenatal care provided by government hospital, transport facility for medical care, baby birth status, mother’s age at first birth, preceding birth interval and vaccination of hepatitis B-1 and B2 are found to be significantly affecting the CS delivery method. Correlation coefficient based PLS regression algorithms may serve more efficiently as a multivariate technique to treat high-dimensional categorical data.

A climate zone approach to global solar radiation modelling using artificial neural networks

Information on solar availability is crucial in the study of both passive building designs and active solar energy systems. There are still many locations within different climate zones worldwide that do not have solar radiation measurements. Correlation between solar radiation and the more commonly measured meteorological variables such as temperature and sunshine hours is useful for locations with no measured solar radiation data. This is also useful for locations with measured solar radiation data, in that any missing data due to equipment breakdown or malfunction can be modelled. Global solar radiation (GSR) was modelled for 96 cities in different climate zones across China using artificial neural networks (ANNs). The novelty of this study is the climate zone approach, by which locations with similar climates were modelled together. Climate classification was based on both the traditional thermal climates and the solar climates. Two sets of models were developed based on measured diurnal temperatures and sunshine hours. Model performance in terms of the predictive power of ANN solar radiation models was evaluated through the Nash-Sutcliffe efficiency coefficient (NSEC). Error analysis of the predicted solar radiation as compared with the measured data was also conducted for each of the 96 cities.

Bayesian hierarchical modeling of the non-stationary traffic conflict extremes for crash estimation

A Bayesian hierarchical modeling (BHM) approach is used to model non-stationary traffic conflict extremes of different sites together for crash estimation. The hierarchical structure has three layers, a data layer that is modeled with a generalized extreme value (GEV) distribution, a latent Gaussian process layer that relates parameters of GEV to covariates and the unobserved heterogeneity, and a prior layer with prior distributions to characterize the latent process. The proposed approach was applied to traffic conflicts collected at the signal cycle level from four intersections in the city of Surrey, British Columbia. Traffic conflicts were measured by the modified time to collision (MTTC) indicator while traffic volume, shock wave area, average shock wave speed, and platoon ration of each cycle were employed as covariates. Four BHM models were developed, including a stationary model (i.e., BHM_GEV(0,0,0)) with no covariates and three non-stationary models (i.e., BHM_GEV(1,0,0), BHM_GEV(0,1,0), and BHM_GEV(1,1,0)) with covariates added to the location parameter, scale parameter, and both parameters of the GEV distribution, respectively. Traditional at-site GEV models were also developed for individual sites for comparison purposes. The results show that the BHM_GEV(1,1,0) is the best fitted model among the four models since considering covariates and unobserved heterogeneity significantly improves the model performance in terms of goodness of fit. The BHM_GEV(1,1,0) also yields relatively accurate and more precise crash estimates compared to the at-site models. This is attributed to the BHM_GEV(1,1,0) allowing borrowing strength from other sites. It is also found that the traffic volume, shock wave area, and platoon ratio have significant influence on the safety of signalized intersections.

Efficient modeling of distributions of polymer properties using probability generating functions and parallel computing

High-fidelity models of polymer processes should include the prediction of distributions of polymer properties, including multivariate distributions. Deterministic models with this capability usually involve a large system of equations, which compromises the model performance in terms of CPU time. The probability generating function (pgf) technique is a powerful method for modeling distributions of polymer properties, including multivariate distributions. It can be applied to systems described by complex kinetic mechanism and requires no a priori assumptions about the distribution shape. The structure of this modeling method makes it particularly suitable for parallel computing. This work describes the application of the pgf technique for modeling uni- and bi-variate distributions of polymer properties with parallelization of the model code. It is shown that accurate results can be achieved in very short running times, which makes the technique suitable for models to be employed in optimization and online control tasks.

A comparative study of concrete strength prediction using artificial neural network, multigene programming and model tree

In the current study 28 day strength of Recycled Aggregate Concrete (RAC) and Fly ash (class F) based concrete is predicted using Artificial Neural Network (ANN), Multigene Genetic Programming (MGGP) and Model Tree (MT). Four sets of models were designed for per cubic proportions of materials, Properties of materials and non-dimensional parameters as input parameters. The study shows that the predicted 28 day strength is in good agreement with the observed data and also generalize well to untrained data. ANN outperforms MGGP and MT in terms of model performance. Output of the developed models can be presented in terms of trained weights and biases in ANN, equations in MGGP and in the form of series of equations in MT. ANN, MGGP and MT can grasp the influence of input parameters which can be seen through Hinton diagrams in ANN, input frequency distribution in MGGP and coefficients of input parameters in MT. The study shows that these data driven techniques can be used for developing model/s to predict strength of concrete with an acceptable performance.

Putting mechanisms in foraging theory: the role of computational mechanisms in optimal decision making

Classic optimization models explain adaptive behaviours in terms of optimal cost–benefit trade-offs. They assume flexibility and make testable predictions about what animals should do in order to maximize their fitness. But they do not usually ask how animals can do it. Since flexible behaviours can be globally optimal but locally suboptimal, the optimality models should directly focus on the underlying mechanisms of behavioural flexibility, such as learning and decision making. In this paper, I use the classic ‘diet’ model of optimal foraging theory (OFT) to investigate the evolution of decision-making mechanisms at both the computational and the algorithmic level. At the computational level, I define benefits (the expected rate of net energy intake) and costs (lost opportunity) and formalize the decision rule. At the algorithmic level, I present two sequential sampling models, which differ in the way information is internally represented and used. The first model represents the prospective items along a one-dimensional scale of values (benefit–cost differences) and it uses a fixed amount of sensory information (and time) to make decisions. The second model represents items in the two-dimensional plane of benefits and costs, and it uses a variable amount of information. I test the models along a gradient of resource abundance. In each environment, I use OFT to classify resources as either profitable or unprofitable and describe the model performance in terms of decision time and accuracy. At high resource density, both models predict foraging choice to be more selective, but less accurate, than at low density, because decisions are strongly biased in the false-positive direction. At low resource density, the two-dimension model performs better than the one-dimension alternative, because it takes less time to make more accurate decisions. These differences, however, disappear when resources are abundant.