Big data analytics: Machine learning and Bayesian learning perspectives—What is done? What is not?

Suthaharan, Shan

doi:10.1002/widm.1283

Cited by 12 publications

(6 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…DNNs with multiple hidden layers, as shown in Appendix A (Figure A1), have a huge number of hyper-parameters (e.g., optimization algorithm, learning rate, network weight initialization, hidden layers activation function, output activation function, L2 regularization, dropout regularization, and the number of nodes in the hidden layers) [87]. The hyper-parameters potentially allow DNNs to perform better in solving the complex problems compared to the other ML models [88]. Sometimes, however, a lack of control over the learning process of the DNNs may lead to overfitting [32].…”

Section: The Individual ML Models In Levelmentioning

confidence: 99%

Improving the Spatial Prediction of Soil Organic Carbon Content in Two Contrasting Climatic Regions by Stacking Machine Learning Models and Rescanning Covariate Space

et al. 2020

View full text Add to dashboard Cite

Understanding the spatial distribution of soil organic carbon (SOC) content over different climatic regions will enhance our knowledge of carbon gains and losses due to climatic change. However, little is known about the SOC content in the contrasting arid and sub-humid regions of Iran, whose complex SOC–landscape relationships pose a challenge to spatial analysis. Machine learning (ML) models with a digital soil mapping framework can solve such complex relationships. Current research focusses on ensemble ML models to increase the accuracy of prediction. The usual ensemble method is boosting or weighted averaging. This study proposes a novel ensemble technique: the stacking of multiple ML models through a meta-learning model. In addition, we tested the ensemble through rescanning the covariate space to maximize the prediction accuracy. We first applied six state-of-the-art ML models (i.e., Cubist, random forests (RF), extreme gradient boosting (XGBoost), classical artificial neural network models (ANN), neural network ensemble based on model averaging (AvNNet), and deep learning neural networks (DNN)) to predict and map the spatial distribution of SOC content at six soil depth intervals for both regions. In addition, the stacking of multiple ML models through a meta-learning model with/without rescanning the covariate space were tested and applied to maximize the prediction accuracy. Out of six ML models, the DNN resulted in the best modeling accuracies, followed by RF, XGBoost, AvNNet, ANN, and Cubist. Importantly, the stacking of models indicated a significant improvement in the prediction of SOC content, especially when combined with rescanning the covariate space. For instance, the RMSE values for SOC content prediction of the upper 0–5 cm of the soil profiles of the arid site and the sub-humid site by the proposed stacking approaches were 17% and 9% respectively, less than that obtained by the DNN models—the best individual model. This indicates that rescanning the original covariate space by a meta-learning model can extract more information and improve the SOC content prediction accuracy. Overall, our results suggest that the stacking of diverse sets of models could be used to more accurately estimate the spatial distribution of SOC content in different climatic regions.

show abstract

Section: The Individual ML Models In Levelmentioning

confidence: 99%

Improving the Spatial Prediction of Soil Organic Carbon Content in Two Contrasting Climatic Regions by Stacking Machine Learning Models and Rescanning Covariate Space

et al. 2020

View full text Add to dashboard Cite

show abstract

“…Bayesian methods for knowledge discovery employ Bayes rule to infer model parameters from big data, construct a probabilistic model of the features and use that model to predict the classification of a newly input dataset (Datcu, Seidel, & Walessa, 1998;Suthaharan, 2019).…”

Section: Basic Principlesmentioning

confidence: 99%

Knowledge discovery from remote sensing images: A review

Wang

Yan

et al. 2020

WIREs Data Min & Knowl

View full text Add to dashboard Cite

The development of Earth observation (EO) technology has made the volume of remote sensing data archiving continually larger, but the knowledge hidden in massive remote sensing images has not been fully exploited. Through indepth research on the artificial intelligence (AI)-based knowledge discovery approaches from remote sensing images, we divided them into four typical types according to their development stage, including rule-based approaches, data-driven approaches, reinforcement learning approaches, and ensemble methods. The basic principles, typical applications, advantages, and disadvantages have been detailed for commonly used algorithms within each category. Conclusions include the following: (a) Rule-based, data-driven and reinforcement learning algorithms form a trilogy from knowledge to data, and to capabilities. (b) Rule-based data mining algorithms can provide prior knowledge for data-driven approaches, the knowledge discovered by data-driven models can be as an important complement to expert knowledge and rule sets, and reinforcement learning approaches can effectively make up for the lack of training samples or small training sample in data-driven models. (c) The traditional data-driven machine learning approaches and their ensemble methods are the current and may be the future mainstream methods for large regional and even global scale long time series remote sensing data mining and analysis, and improving their computing efficiency is the key research direction.(d) Deep learning, deep reinforcement learning, transfer learning, and an ensemble approach of the three may be the main means for small-area scope, short time series, and key geoscience information extraction from remote sensing images within a long time of the future.

show abstract

“…Challenges, methodologies and applications related to industrial big data analytics was presented in [19]. Machine learning and Bayesian learning perspectives to big data characterizing data heterogeneity was presented in [20]. Though the above said methods, ensured dimensionality reduction in the perspective of big data, less focus was made on the performance of classification accuracy and classification time during prediction process with data warehouse and big data.…”

Section: Related Workmentioning

confidence: 99%

Tobit Regressive Based Gaussian Independence Bayes Map Reduce Classifier on Data Warehouse for Predictive Analytics

Sivakkolundu¹,

Kavitha²

2019

IJITEE

View full text Add to dashboard Cite

Data warehouse comprises of data collected from different probable heterogeneous resources at different time intervals with the objective of responding to user analytic queries. Big data is a field that helps in analysing and extracting information from large datasets. The unfolding Big Data incorporation inflicts multiple confronts, compromising the feasible business research practice. Heterogeneous resources, high dimensionality and massive volumes that confront Big Data prototype may prevent the effectual data and system integration processes. In this work, we plan to develop a Tobit Regressive based Gaussian Independence Bayes Map Reduce Classifier (TR-GIBMRC) method for categorizing the collected and stored data which helps the users in making decision with minimum time consumption. The TR-GIBMRC method consists of two processes. They are, Tobit Regressive Feature Selection and Gaussian Independence Bayes Map Reduce Classification. Tobit Regressive Feature Selection process is used to select relevant features from collected and stored data. Tobit statistical model, used to describe the relationship between non-negative dependent variable and an independent variable for selecting relevant features. Next, Gaussian Independence Bayes Map Reduce Classifier is used to classify the selected relevant features for decision making with lesser time consumption. Gaussian Independence Bayes Map Reduce Classifier, a probabilistic classifier segments the data by class by measuring the mean and variance of data in each class. The data point gets allocated to the class with minimal variance. This in turn helps to perform efficient data classification for accurate decision making. Experimental evaluation is carried out on the factors such as feature selection rate, classification accuracy, classification time and error rate with respect to number of features and number of data points.

show abstract

Big data analytics: Machine learning and Bayesian learning perspectives—What is done? What is not?

Cited by 12 publications

References 31 publications

Improving the Spatial Prediction of Soil Organic Carbon Content in Two Contrasting Climatic Regions by Stacking Machine Learning Models and Rescanning Covariate Space

Improving the Spatial Prediction of Soil Organic Carbon Content in Two Contrasting Climatic Regions by Stacking Machine Learning Models and Rescanning Covariate Space

Knowledge discovery from remote sensing images: A review

Tobit Regressive Based Gaussian Independence Bayes Map Reduce Classifier on Data Warehouse for Predictive Analytics

Contact Info

Product

Resources

About