Grid Search Approach Research Articles

A Comparison between Three Tuning Strategies for Gaussian Kernels in the Context of Univariate Genomic Prediction.

Genomic prediction is revolutionizing plant breeding since candidate genotypes can be selected without the need to measure their trait in the field. When a reference population contains both phenotypic and genotypic information, it is trained by a statistical machine learning method that is subsequently used for making predictions of breeding or phenotypic values of candidate genotypes that were only genotyped. Nevertheless, the successful implementation of the genomic selection (GS) methodology depends on many factors. One key factor is the type of statistical machine learning method used since some are unable to capture nonlinear patterns available in the data. While kernel methods are powerful statistical machine learning algorithms that capture complex nonlinear patterns in the data, their successful implementation strongly depends on the careful tuning process of the involved hyperparameters. As such, in this paper we compare three methods of tuning (manual tuning, grid search, and Bayesian optimization) for the Gaussian kernel under a Bayesian best linear unbiased predictor model. We used six real datasets of wheat (Triticum aestivum L.) to compare the three strategies of tuning. We found that if we want to obtain the major benefits of using Gaussian kernels, it is very important to perform a careful tuning process. The best prediction performance was observed when the tuning process was performed with grid search and Bayesian optimization. However, we did not observe relevant differences between the grid search and Bayesian optimization approach. The observed gains in terms of prediction performance were between 2.1% and 27.8% across the six datasets under study.

An Insight into the Prediction of Scale Precipitation in Harsh Conditions Using Different Machine Learning Algorithms

Summary Scale precipitation in petroleum equipment is known as an important problem that causes damages in injection and production wells. Scale precipitation causes equipment corrosion and flow restriction and consequently a reduction in oil production. Due to this fact, the prediction of scale precipitation has vital importance among petroleum engineers. In the current work, different intelligent models, including the decision tree, random forest (RF), artificial neural network (ANN), K-nearest neighbors (KNN), convolutional neural network (CNN), support vector machine (SVM), ensemble learning, logistic regression, Naïve Bayes, and adaptive boosting (AdaBoost), are used to estimate scale formation as a function of pH and ionic compositions. Also, a sensitivity analysis is done to determine the most influential parameters on scale formation. The novelty of this work is to compare the performance of 10 different machine learning algorithms at modeling an extremely non-linear relationship between the inputs and the outputs in scale precipitation prediction. After determining the best models, they can be used to determine scale formation by manipulating the concentration of a variable in accordance with the result of the sensitivity analysis. Different classification metrics, including the accuracy, precision, F1-score, and recall, were used to compare the performance of the mentioned models. Results in the testing phase showed that the KNN and ensemble learning were the most accurate tools based on all performance metrics of solving the classification of scale/no-scale problem. As the output had an extremely non-linear behavior in terms of the inputs, an instance-based learning algorithm such as the KNN best suited the classification task in this study. This argumentation was backed by the classification results. Furthermore, the SVM, Naïve Bayes, and logistic regression performance metrics were not satisfactory in the prediction of scale formation. Note that the hyperparameters of the models were found by grid search and random search approaches. Finally, the sensitivity analysis showed that the variations in the concentration of Ca had the highest impact on scale precipitation.

DLCP2F: a DL-based cryptocurrency price prediction framework

Cryptocurrencies are distributed digital currencies that have emerged as a consequence of financial technology advancement. In 2017, cryptocurrencies have shown a huge rise in their market capitalization and popularity. They are now employed in today’s financial systems as individual investors, corporate firms, and big institutions are heavily investing in them. However, this industry is less stable than traditional currency markets. It can be affected by several legal, sentimental, and technical factors, so it is highly volatile, dynamic, uncertain, and unpredictable, hence, accurate forecasting is essential. Recently, cryptocurrency price prediction becomes a trending research topic globally. Various machine and deep learning algorithms, e.g., Neural Networks (NN), Gated Recurrent Unit (GRU), Long Short-Term Memory (LSTM), and Bidirectional LSTM (BiLSTM) were utilized to analyze the factors influencing the prices of the cryptocurrencies and accordingly predict them. This paper suggests a five-phase framework for cryptocurrency price prediction based on two state-of-the-art deep learning architectures (i.e., BiLSTM and GRU). The current study uses three public real-time cryptocurrency datasets from “Yahoo Finance”. Bidirectional Long Short-Term Memory and Gated Recurrent Unit deep learning-based algorithms are used to forecast the prices of three popular cryptocurrencies (i.e., Bitcoin, Ethereum, and Cardano). The Grid Search approach is used for the hyperparameters optimization processes. Results indicate that GRU outperformed the BiLSTM algorithm for Bitcoin, Ethereum, and Cardano, respectively. The lowest RMSE for the GRU model was found to be 0.01711, 0.02662, and 0.00852 for Bitcoin, Ethereum, and Cardano, respectively. Experimental results proved the significant performance of the proposed framework that achieves the minimum MSE and RMSE values.

Sentiment Analysis on Movie Reviews Dataset Using Support Vector Machines and Ensemble Learning

The internet makes it easier for people to connect to each other and has become a platform to express ideas and share information with the world. The growth of the internet has indirectly led to the development of social networking sites. The reviews posted by people on these sites implies their opinion, and analysis over reviews is required to understand their intent. In this paper, natural language processing technique and machine learning algorithms are applied to classify the text data. The contributions of the proposed approach are three-fold: 1) chi square selector is applied to select the k-best features, 2) support vector machines is executed to classify the reviews (hyperparameters of the SVM classifier are tuned using GridSearch approach), and 3) bagging algorithm is applied with the base classifier over the newly built SVM classifier. The number of base classifiers of the bagging algorithm is varied accordingly. The results of the proposed approach are compared to the similar existing work, and hence, it is found to achieve better results as compared to the existing systems.

Prediction of depression among patients with cardiovascular disease scheduled cardiac surgery using an AI-empowered pipeline

Abstract Introduction Several studies have demonstrated, that one in three patients with cardiovascular disease (CVD) suffers from depression, and depression increases the likelihood for cardiac morbidity and mortality in the CVD population by 2–3-fold, independently of traditional risk factors or gender. Purpose The aim of this work is to predict depression in patients with CVD. Methods The clinical study was conducted in the cardiac surgical intensive care unit of The Central Clinical Hospital, the Medical University of Lodz in Poland, among patients scheduled for elective coronary artery bypass graft surgery (CABG). 224 patients signed an informed consent form, met the inclusion criteria and were enrolled in the study. All the patients had chronic coronary syndromes. The inclusion criteria were: consecutive adult patients scheduled for CABG surgery or CABG surgery with cardiac valve repair or replacement (CVR). The study population was examined by a psychiatrist the day prior to the scheduled operation, a diagnosis of major depressive disorder (MDD) was established on the basis of DSM-5 criteria. A data curation pipeline was applied to automatically remove outliers and duplicated fields in the input dataset. An AI-empowered pipeline was developed to classify patients at higher risk for depression. Random downsampling with replacement was applied to deal with the increased class imbalance by taking into consideration two confound factors, namely the Mini Mental State Examination and Hemoglobin concentration <10mg/dl. The downsampling process was repeated K times. In each iteration, three bagging and boosting ML schemas were utilized for the classification task including the AdaBoost (adaptive boosting), Random Forests and Extreme Gradient boosting trees (XGBoost). To this end, a nested cross-validation process was applied for hyperparameter optimization and model validation, where: (i) a 3-fold cross-validation process was first applied to seek for the optimal set of hyperparameters based on the grid search approach by tuning core parameters, including the learning rate, number of estimators, and max depth, among others, and (ii) a stratified 5-fold cross-validation process was subsequently used to evaluate the performance of the best model from the previous stage by computing the classification accuracy, sensitivity, specificity, and area under the ROC curve (AUC). The performance evaluation results were averaged across the 5 folds and across the downsampling iterations. Shappley additive explanation analysis was finally applied to provide explainable risk factors for depression. Results sRAGE was significantly correlated with depression (r=0.32, p=0.038). The Random forests classifier presented the highest performance to predict depression (Accuracy: 0.71, Sensitivity: 0.71, Specificity: 0.75 and AUC: 0.74). Conclusions Depression can be predicted with 71% accuracy at patients with CVD and elevated sRAGE values. Funding Acknowledgement Type of funding sources: Public grant(s) – EU funding. Main funding source(s): HORIZON2020

Interpretable machine learning methods to explain on-farm yield variability of high productivity wheat in Northwest India

The increasing availability of complex, geo-referenced on-farm data demands analytical frameworks that can guide crop management recommendations. Recent developments in interpretable machine learning techniques offer opportunities to use these methods in agronomic studies. Our objectives were two-fold: (1) to assess the performance of different machine learning methods to explain on-farm wheat yield variability in the Northwestern Indo-Gangetic Plains of India, and (2) to identify the most important drivers and interactions explaining wheat yield variability. A suite of fine-tuned machine learning models (ridge and lasso regression, classification and regression trees, k-nearest neighbor, support vector machines, gradient boosting, extreme gradient boosting, and random forest) were statistically compared using the R 2 , root mean square error (RMSE), and mean absolute error (MAE). The best performing model was again fine-tuned using a grid search approach for the bias-variance trade-off. Three post-hoc model agnostic techniques were used to interpret the best performing model: variable importance (a variable was considered “important” if shuffling its values increased or decreased the model error considerably), interaction strength (based on Friedman’s H-statistic), and two-way interaction (i.e., how much of the total variability in wheat yield was explained by a particular two-way interaction). Model outputs were compared against empirical data to contextualize results and provide a blueprint for future analysis in other production systems. Tree-based and decision boundary-based methods outperformed regression-based methods in explaining wheat yield variability. Random forest was the best performing method in terms of goodness-of-fit and model precision and accuracy with RMSE, MAE, and R 2 ranging between 367 and 470 kg ha −1 , 276–345 kg ha −1 , and 0.44–0.63, respectively. Random forest was then used for selection of important variables and interactions. The most important management variables explaining wheat yield variability were nitrogen application rate and crop residue management, whereas the average of monthly cumulative solar radiation during February and March (coinciding with reproductive phase of wheat) was the most important biophysical variable. The effect size of these variables on wheat yield ranged between 227 kg ha −1 for nitrogen application rate to 372 kg ha −1 for cumulative solar radiation during February and March. The effect of important interactions on wheat yield was detected in the data namely the interaction between crop residue management and disease management and, nitrogen application rate and seeding rate. For instance, farmers’ fields with moderate disease incidence yielded 750 kg ha −1 less when crop residues were removed than when crop residues were retained. Similarly, wheat yield response to residue retention was higher under low seed and N application rates. As an inductive research approach, the appropriate application of interpretable machine learning methods can be used to extract agronomically actionable information from large-scale farmer field data. • Data-driven agronomic research requires new analytical and methodological approaches. • Machine learning methods were used to disentangle complex relationships in farmer field data. • Model-agnostic tools were used to derive agronomic interpretations. • Residue management and N application rate were important management variables for wheat yield. • Residue management interacted with other practices to explain wheat yield variability.

Heat energy consumption forecast analysis: model-oriented multi-objective approach

Forecasting the expected quantity of heat energy consumption by multiple residential buildings can be a time consuming task, if done manually. The energy consumption in a building can be represented as a data parameter that changes over a period of time. For forecasting purposes autoregressive models are most often used to analyse time series data. One solution of the problem is to perform a hyperparameter optimization and derive the best results according to a selected metric. However, the grid search approach is often a time-consuming and resource-intensive task requiring additional human interaction for the final assessment of the most suitable parameters. The purpose of this study is to explore the possibility of developing a fully automated software algorithm that will provide sufficient parameters for the prediction of heat energy using the seasonal autoregressive integrated moving average with exogenous variables model (SARIMAX).

A Topological Machine Learning Pipeline for Classification

In this work, we develop a pipeline that associates Persistence Diagrams to digital data via the most appropriate filtration for the type of data considered. Using a grid search approach, this pipeline determines optimal representation methods and parameters. The development of such a topological pipeline for Machine Learning involves two crucial steps that strongly affect its performance: firstly, digital data must be represented as an algebraic object with a proper associated filtration in order to compute its topological summary, the Persistence Diagram. Secondly, the persistence diagram must be transformed with suitable representation methods in order to be introduced in a Machine Learning algorithm. We assess the performance of our pipeline, and in parallel, we compare the different representation methods on popular benchmark datasets. This work is a first step toward both an easy and ready-to-use pipeline for data classification using persistent homology and Machine Learning, and to understand the theoretical reasons why, given a dataset and a task to be performed, a pair (filtration, topological representation) is better than another.

Thin-Film Batteries Failure Prediction from a Short Pulsed Test Using Machine Learning

We experimentally demonstrated for the first time in 2019 LiCoO2 all solid-state thin-film batteries with high electrochemical density storage of 0.89 mA.h.cm-2 [1]. Despite these excellent performances, some devices may be defective and fail during cycling. Therefore, in the context of industrial production of such devices, there is a need to reliably assess the functionality of the cells on the wafer scale. The most straightforward approach consists in simply cycling the batteries as well as performing complementary short tests such as Open-Circuit Voltage measurement (OCV) and Electrochemical Impedance Spectroscopy (EIS). However, with 14 000+ cells per wafer said approach is too time consuming. Moreover tested batteries would not be suitable for further commercialization as the electrodes have already been solicited, which could affect their shelf life.We therefore have investigated on the ability of a short – i.e. with a duration in the order of the minute – pulsed test to predict the failure of our dies, using machine learning. This Electrical Wafer Sorting (EWS) or screening protocol consists in inducing a partial charge for 1 min and then to observe the relaxation by measuring the OCV for 10 s. OCV and EIS are also measured before and after the pulse. Certain features are extracted from profiles collected during the EWS protocol (shown in Figure a)), such as the mean OCV or the real part of the impedance at 1 kHz. 1122 devices were screened using this protocol and then cycled for 1 cycle (1 full charge 1 full discharge). The batteries were then classified in binary categories, according to the outcome of the cycling test: fail or pass. Fail category encompasses all failure modes: (e.g. open-circuit, short-circuit). The features extracted during the EWS were used to build the prediction model, whose target is the cycling categorization. Foremost, the data was split in a training set and a test set. In a first step, variable selection was achieved using a simple decision tree model. It enabled us to choose the combination of parameters leading to a maximization of the f1-scores (from 68 to 73 %). However, the learning curves revealed that this simple model clearly overfitted and was unable to generalize. In a second step, a variety of models were trained using the selected features, including Random Forest, Support Vector Machines, and Logistic Regression. The three models leading to best f1-scores were kept and their hyperparameters have been further optimized following a grid search approach. The algorithm giving the highest scores is the Random Forest classifier. Although it was not possible to increase its accuracy (still at 73%) the optimization step helped to significantly reduce the overfitting, reinforcing the ability of the model to be equally accurate on new data. The corresponding confusion matrix and learning curves are displayed on Figure b). Figure c) shows respectively predicted and actual wafermaps for data present in the dataset used to build the model. An example of predicted wafermap on a complete wafer is presented on Figure d).Achieved performance provide an interesting base for investigations in a research context. It is shown that the functionality of the thin film batteries can be predicted with reasonable confidence. Nevertheless, far better scores (above 90%) are required for production application. Thus further work will focus on collecting new data to check on generalizability and if it helps to improve the model accuracy when used for training. All the work presented in this abstract has been carried out using scikit-learn Python package [2].[1] S. Oukassi et al., “Millimeter scale thin film batteries for integrated high energy density storage,” 2019, vol. 2019-December. doi: 10.1109/IEDM19573.2019.8993483.[2] F. Pedregosa et al., “Scikit-learn: Machine Learning in Python,” Journal of Machine Learning Research, vol. 12, no. 85, pp. 2825–2830, 2011. Figure 1

A Deep Learning Approach-FDNN: Forest Deep Neural Network to Predict Cow’s Parturition Date

In this prospective study, we integrate neural network architecture with a supervised random forest feature detector to develop a new model named Forest Deep Neural Network (FDNN) to predict daily and hourly calving time of cattle. To overcome challenges of prediction problems like data sparsity along with unknown correlation structures, we incorporate the benefits of random forest (RF) with a deep neural network (DNN) to predict the daily and hourly calving time of cattle, which is nobody done yet. For this study, we take a total of 45 Holstein-Friesian cows (27 primiparous and 18 multiparous) for collecting physical activities. Using IceQube and HR Tag technologies, we record daily and hourly lying time, the number of stand-ups, ruminating time, the number of steps, and the number of head moves of cattle from 15 days before the actual calving time. Different statistical analysis has been carried out over the daily and hourly-captured data, and we have found that these monitored physical activities change very significantly over time. We have applied five classifiers such as FDNN, DNN, RF, decision tree (DT), and support vector machine (SVM) over the daily and hourly datasets. Hyperparameter optimization has been conducted over the classifiers using Grid Search approach to filter out the optimal parameter configurations. With optimal parameters, our developed model overpowered the other four classifiers in terms of accuracy, sensitivity, specificity, and ROC score (ACC= 98.38, SN=88.19, SP=98.41, and ROC=99 of predicting daily calving time; ACC=97.93, SN=97.40, SP=89.42, and ROC=98 of predicting hourly calving time).

Optimization of RFM for automated breast cancer detection

Recent years have seen an upsurge in the acceptance of illness diagnosis and prediction utilizing ML algorithms. A ML model can be employed in the diagnosis of breast cancer illness. In this research, an effective breast cancer prediction model with grid search approach is provided. Using the random forest approach, grid search is used to find the best n-estimator, which may provide the highest possible accuracy for predicting breast cancer. The accuracy of the suggested model can then be utilised to contrast its effectiveness to that of a standard RFM. The experimental result analysis demonstrates that the optimized model has 97.07 percent accuracy whereas the regular random forest technique has an accuracy of 94.73 percent in breast cancer detection.

Lowermost mantle (D″ layer) structure beneath the Indian Ocean: Insights from modeling of ScS-S and PcP-P residuals

• High Compressional (P) and shear (S) wave velocities in the lowermost mantle ( D ″ layer) beneath the Indian Ocean Geoid Low. • Evidence for slabs at the CMB beneath Indian Ocean Geoid Low. • Low R values in the lowermost mantle ( D ″ layer) beneath the western Indian Ocean Geoid Low. In this study, we investigate the seismic structure of the D ″ layer beneath the Indian Ocean by modeling the ScS-S and PcP-P differential travel time residuals corrected for the velocity structure above it. These times, representative of the anomalies in the D ″ layer, vary from −9.58 to 5.06 s for the shear wave (ScS-S) and −4.54 to 3.14 s for the compressional wave (PcP-P). Modeling of the residuals using a grid search approach reveals velocity perturbations in the range of −3.06% to 5.72% for the shear and −4.81% to 5.47% for the compressional waves, in the D ″ layer. Interestingly, the perturbations are positive below the Indian Ocean Geoid Low (IOGL) and negative below the adjoining region. The results clearly reveal presence of high velocity material atop the Core Mantle Boundary (CMB) beneath the IOGL, representing dehydrated Tethyan subducted slabs. Further, the low R SP values in the lowermost mantle beneath the western part of IOGL, calculated using the P and S velocity estimates from this study, mostly lie between 0.01 and 2.7. This implies that the anomalies may be thermal in origin, owing to the heterogeneities resulting from cold lithospheric slabs at the CMB.

Integrative artificial intelligence models for Australian coastal sediment lead prediction: An investigation of in-situ measurements and meteorological parameters effects

Heavy metals (HMs) such as Lead (Pb) have played a vital role in increasing the sediments of the Australian bay's ecosystem. Several meteorological parameters (i.e., minimum, maximum and average temperature (Tmin, Tmax and TavgoC), rainfall (Rn mm) and their interactions with the other batch HMs, are hypothesized to have high impact for the decision-making strategies to minimize the impacts of Pb. Three feature selection (FS) algorithms namely the Boruta method, genetic algorithm (GA) and extreme gradient boosting (XGBoost) were investigated to select the highly important predictors for Pb concentration in the coastal bay sediments of Australia. These FS algorithms were statistically evaluated using principal component analysis (PCA) Biplot along with the correlation metrics describing the statistical characteristics that exist in the input and output parameter space of the models. To ensure a high accuracy attained by the applied predictive artificial intelligence (AI) models i.e., XGBoost, support vector machine (SVM) and random forest (RF), an auto-hyper-parameter tuning process using a Grid-search approach was also implemented. Cu, Ni, Ce, and Fe were selected by all the three applied FS algorithms whereas the Tavg and Rn inputs remained the essential parameters identified by GA and Boruta. The order of the FS outcome was XGBoost > GA > Boruta based on the applied statistical examination and the PCA Biplot results and the order of applied AI predictive models was XGBoost-SVM > GA-SVM > Boruta-SVM, where the SVM model remained at the top performance among the other statistical metrics. Based on the Taylor diagram for model evaluation, the RF model was reflected only marginally different so overall, the proposed integrative AI model provided an evidence a robust and reliable predictive technique used for coastal sediment Pb prediction.

Modelling the gamma-ray morphology of HESSJ1804−216 from two supernova remnants in a hadronic scenario

ABSTRACT HESS J1804−216 is one of the brightest yet most mysterious TeV γ-ray sources discovered to date. Previous arc-minute scale studies of the interstellar medium (ISM) surrounding this TeV γ-ray source revealed HESS J1804−216 is likely powered by a mature supernova remnant (SNR) or pulsar, hence its origin remains uncertain. In this paper, we focus on the diffusive escape of cosmic ray protons from potential SNR accelerators. These cosmic rays interact with the ISM to produce TeV γ-rays. We utilize the isotropic diffusion equation solution for particles escaping from a shell, to model the energy-dependent escape and propagation of protons into the ISM. This work is the first attempt at modelling the spatial morphology of γ-rays towards HESS J1804−216, using arc-minute ISM observations from both Mopra and the Southern Galactic Plane Survey. The spectral and spatial distributions of γ-rays for the two nearby potential SNR counterparts, SNR G8.7−0.1 and the progenitor SNR of PSR J1803−2137, are presented here. We vary the diffusion parameters and particle spectrum and use a grid search approach to find the best combination of model parameters. We conclude that moderately slow diffusion is required for both candidates. The most promising candidate to be powering the TeV γ-rays from HESS J1804−216 in a hadronic scenario is the progenitor SNR of PSR J1803−2137.

A machine-learning–based algorithm improves prediction of preeclampsia-associated adverse outcomes

Preeclampsia presents a highly prevalent burden on pregnant women with an estimated incidence of 2% to 5%. Preeclampsia increases the maternal risk of death 20-fold and is one of the main causes of perinatal morbidity and mortality. Novel biomarkers, such as soluble fms-like tyrosine kinase-1 and placental growth factor in addition to a wide span of conventional clinical data (medical history, physical symptoms, laboratory parameters, etc.), present an excellent basis for the application of early-detection machine-learning models. This study aimed to develop, train, and test an automated machine-learning model for the prediction of adverse outcomes in patients with suspected preeclampsia. Our real-world dataset of 1647 (2472 samples) women was retrospectively recruited from women who presented to the Department of Obstetrics at the Charité - Universitätsmedizin Berlin, Berlin, Germany, between July 2010 and March 2019. After standardization and data cleaning, we calculated additional features regarding the biomarkers soluble fms-like tyrosine kinase-1 and placental growth factor and sonography data (umbilical artery pulsatility index, middle cerebral artery pulsatility index, mean uterine artery pulsatility index), resulting in a total of 114 features. The target metric was the occurrence of adverse outcomes throughout the remaining pregnancy and 2 weeks after delivery. We trained 2 different models, a gradient-boosted tree and a random forest classifier. Hyperparameter training was performed using a grid search approach. All results were evaluated via a 10× 10-fold cross-validation regimen. We obtained metrics for the 2 naive machine-learning models. A gradient-boosted tree model was performed with a positive predictive value of 88%±6%, a negative predictive value of 89%±3%, a sensitivity of 66%±5%, a specificity of 97%±2%, an overall accuracy of 89%±3%, an area under the receiver operating characteristic curve of 0.82±0.03, an F1 score of 0.76±0.04, and a threat score of 0.61±0.05. The random forest classifier returned an equal positive predictive value (88%±6%) and specificity (97%±1%) while performing slightly inferior on the other available metrics. Applying differential cutoffs instead of a naive cutoff for positive prediction at ≥0.5 for the classifier's results yielded additional increases in performance. Machine-learning techniques were a valid approach to improve the prediction of adverse outcomes in pregnant women at high risk of preeclampsia vs current clinical standard techniques. Furthermore, we presented an automated system that did not rely on manual tuning or adjustments.

Accelerating online learning: Machine learning insights into the importance of cumulative experience, independence, and country setting

Cumulative experience is important for developing expertise through in-person learning, along with country setting and gender, but evidence is limited the role of these features in online learning. Yet, COVID-19 has catalysed the centrality of online learning, such that the efficacy of online learning is now highly relevant. Although the Pandemic triggered a surge of self-report and literature review research on stakeholder perceptions of online learning, less educational research has used big data to understand online learning. Therefore, the present research mined online learning data to identify features that are important for developing expertise in online learning. Data mining of 54,842,787 initial data points from one online learning platform was conducted by partnering theory with data in model development. Following examination of a theory-led machine learning model, a data-led approach was taken to reach a final model. The linear regression model was regularised with the Lasso penalty to enable data-driven feature selection. Twenty-six features were selected to form an extreme gradient boosting model that underwent hyper-parameter tuning. All cross-validation adopted the grid search approach. The final model was used to derive Shapley values for feature importance. As expected, cumulative experience, country differences, low-and-middle-income country status, and COVID-19 were important features for developing expertise through online learning. The data-led model development resulted in additional insights not examined in the initial, theory-led model: namely, the importance of meta-cognition and independent learner behaviour. Surprisingly, no male advantage was found in the potential for expertise development through online learning. • Machine learning uncovered theoretically relevant insights into online learning. • Cumulative experience was central to developing expertise through online learning. • Independent rather than help-seeking learners made the most academic gains. • Countries differed in the potential for expertise development in online learning. • COVID-19 may have improved the efficacy of developing expertise via online learning.

Mapping Invasive Plant Species with Hyperspectral Data Based on Iterative Accuracy Assessment Techniques

Recent developments in computer hardware made it possible to assess the viability of permutation-based approaches in image classification. Such approaches sample a reference dataset multiple times in order to train an arbitrary number of machine learning models while assessing their accuracy. So-called iterative accuracy assessment techniques or Monte-Carlo-based approaches can be a useful tool when it comes to assessment of algorithm/model performance but are lacking when it comes to actual image classification and map creation. Due to the multitude of models trained, one has to somehow reason which one of them, if any, should be used in the creation of a map. This poses an interesting challenge since there is a clear disconnect between algorithm assessment and the act of map creation. Our work shows one of the ways this disconnect can be bridged. We calculate how often a given pixel was classified as given class in all variations of a multitude of post-classification images delivered by models trained during the iterative assessment procedure. As a classification problem, a mapping of Calamagrostis epigejos, Rubus spp., Solidago spp. invasive plant species using three HySpex hyperspectral datasets collected in June, August and September was used. As a classification algorithm, the support vector machine approach was chosen, with training hyperparameters obtained using a grid search approach. The resulting maps obtained F1-scores ranging from 0.87 to 0.89 for Calamagrostis epigejos, 0.89 to 0.97 for Rubus spp. and 0.99 for Solidago spp.

Best phase-forecasting method for brain-state dependent stimulation: A grid-search approach

Best phase-forecasting method for brain-state dependent stimulation: A grid-search approach

Novel dual-channel long short-term memory compressed capsule networks for emotion recognition

Recent analysis on speech emotion recognition (SER) has made considerable advances with the use of MFCC’s spectrogram features and the implementation of neural network approaches such as convolutional neural networks (CNNs). The fundamental issue of CNNs is that the spatial information is not recorded in spectrograms. Capsule networks (CapsNet) have gained gratitude as alternatives to CNNs with their larger capacities for hierarchical representation. However, the concealed issue of CapsNet is the compression method that is employed in CNNs cannot be directly utilized in CapsNet. To address these issues, this research introduces a text-independent and speaker-independent SER novel architecture, where a dual-channel long short-term memory compressed-CapsNet (DC-LSTM COMP-CapsNet) algorithm is proposed based on the structural features of CapsNet. Our proposed novel classifier can ensure the energy efficiency of the model and adequate compression method in speech emotion recognition, which is not delivered through the original structure of a CapsNet. Moreover, the grid search (GS) approach is used to attain optimal solutions. Results witnessed an improved performance and reduction in the training and testing running time. The speech datasets used to evaluate our algorithm are: Arabic Emirati-accented corpus, English “speech under simulated and actual stress (SUSAS)” corpus, English Ryerson audio-visual database of emotional speech and song (RAVDESS) corpus, and crowd-sourced emotional multimodal actors dataset (CREMA-D). This work reveals that the optimum feature extraction method compared to other known methods is MFCCs delta-delta. Using the four datasets and the MFCCs delta-delta, DC-LSTM COMP-CapsNet surpasses all the state-of-the-art systems, classical classifiers, CNN, and the original CapsNet. Using the Arabic Emirati-accented corpus, our results demonstrate that the proposed work yields average emotion recognition accuracy of 89.3% compared to 84.7%, 82.2%, 69.8%, 69.2%, 53.8%, 42.6%, and 31.9% based on CapsNet, CNN, support vector machine (SVM), multi-layer perceptron (MLP), k-nearest neighbor (KNN), radial basis function (RBF), and naïve Bayes (NB), respectively.

PSXV-9 Transgenerational epigenetic variance for production and reproduction traits in maternal-line pigs

Abstract Modeling epigenetic factors impacting phenotypic expression of economically important traits has become a hot-topic in the field of animal breeding due to the variability in genetic expression caused by environmental stressors (e.g., heat stress). This variability may be due, in part, to in-utero epigenomic remodeling, which has been reported to be passed from parent to offspring. We aimed to estimate transgenerational epigenetic variance for various production and reproduction traits measured in a maternal-line pig population, using a Bayesian approach. The phenotypes for production [n = 10,862; i.e., weaning weight (WW), birth weight (BW) and ultrasound-backfat thickness (BF)] and reproduction [n = 5,235, i.e., number of piglets born alive (NBA) and total number of piglets born (TB)] traits from a purebred Landrace population were provided by Smithfield Premium Genetics (NC, USA). The pedigree information traced back to 10 generations. Single-trait genetic analyses were performed using mixed models that included additive genetic, common environmental, and epigenetic random effects. The Gibbs sampler algorithm based on Markov chain Monte Carlo was used to estimate the variance components. The epigenetic relationship matrix was constructed using a recursive parameter (λ) related to the transmissibility coefficient of epigenetic markers. A grid search approach was used to define the optimal λ value (λ values ranged from 0.1 to 0.5, with an interval of 0.1). The optimal λ value was determined based on the deviance information criterion, and it was used to estimate the additive and epigenetic variances. For instance, based on preliminary results, the optimal λ value estimated for TB was 0.3 with an additive genetic variance of 0.94 (0.19 PSD) and epigenetic variance of 0.67 (0.18 PSD). The additive genetic heritability was 0.076 (0.015 PSD) and the estimated epigenetic heritability was 0.053 (0.015 PSD). This preliminary result suggests that epigenetics contribute to the non-Mendelian variability in pigs.