An assessment of ensemble learning approaches and single-based machine learning algorithms for the characterization of undersaturated oil viscosity

Abstract

Background Prediction of accurate crude oil viscosity when pressure volume temperature (PVT) experimental results are not readily available has been a major challenge to the petroleum industry. This is due to the substantial impact an inaccurate prediction will have on production planning, reservoir management, enhanced oil recovery processes and choice of design facilities such as tubing, pipeline and pump sizes. In a bid to attain improved accuracy in predictions, recent research has focused on applying various machine learning algorithms and intelligent mechanisms. In this work, an extensive comparative analysis between single-based machine learning techniques such as artificial neural network, support vector machine, decision tree and linear regression, and ensemble learning techniques such as bagging, boosting and voting was performed. The prediction performance of the models was assessed by using five evaluation measures, namely mean absolute error, relative squared error, mean squared error, root mean squared error and root mean squared log error. Results The ensemble methods offered generally higher prediction accuracies than single-based machine learning techniques. In addition, weak single-based learners of the dataset used in this study (for example, SVM) were transformed into strong ensemble learners with better prediction performance when used as based learners in the ensemble method, while other strong single-based learners were discovered to have had significantly improved prediction performance. Conclusion The ensemble methods have great prospects of enhancing the overall predictive accuracy of single-based learners in the domain of reservoir fluid PVT properties (such as undersaturated oil viscosity) prediction.

Comparison of Bagging, Boosting, and Stacking Ensemble Models for Airline Customer Satisfaction Analysis

“…Past studies on analyses of a multidimensional problem have seen higher prediction accuracy using ensemble learning than single-based machine learning techniques (Akano & James, 2022), with many literature reviews on various ensemble learning techniques (Dong et al, 2020). However, despite the strengths, ensemble learning process still have its weaknesses to be aware of, which despite several strategies and techniques still have limitations in terms of generalization, training difficulties, and more (Tasci et al, 2021).…”

Improving the Quality Indicators of Multilevel Data Sampling Processing Models Based on Unsupervised Clustering

“…Aggregating several basic models provides an opportunity to smooth out the deviations of each algorithm when external influences occur. In transforming the properties of the processed information sequence, the problem of determining noise, outliers, or detecting changes in data properties arises, leading to significant resource consumption for the retraining of classification algorithms [5]. Assigning a separate model to a data segment makes it possible to reduce the labor and resource intensity of these processes and increases the speed of model adaptation to the changed properties of the input sequences of observation objects.…”

“…Its manifestation is characteristic of using both complex models and a large number of relatively simple data processing algorithms [6]. Ensemble retraining is a more costly procedure than individual model retraining [5,7]. Combining multiple models requires solving several problematic issues related to their settings to ensure "distinguishability" in the case of large datasets or limited computational resources [2].…”

“…The papers provide various ways of combining ensemble models, but their selection and building are empirical, which does not make it possible to transfer the experience to other subject areas and data structures [2,3]. Another problem with ensemble approaches is that setting optimal parameters requires considerable knowledge and experience and many analysis procedures [4,5]. Combining the results of different classification algorithms cannot always reduce the processing error [4,6].…”

“…In any case, the ensemble quality usually depends on the quality of the underlying models. However, the aggregation functions used in ensembles do not always allow us to respond quickly to changes in data properties related to changes in the distribution, frequency of occurrence of observation objects, presence of noise, and outliers, according to Mohammed & Kora (2023) [2] and Akano & James (2022) [5].…”