Abstract
The rapid growth of e-commerce platforms presents unique opportunities to analyze consumer behavior and predict product preferences in the furniture industry. This study explores the use of machine learning techniques to predict consumer choices for furniture products based on favorite listing data from e-commerce platforms. A dataset of 239 furniture products was collected, categorized into three groups: most preferred, moderately preferred, and least preferred. Key attributes, including furniture type, dimensions (width, depth, height), color, material, and price, were analyzed. Machine learning models, specifically Decision Trees and Random Forests, were applied to develop prediction models for these categories. The models were assessed using metrics such as accuracy, precision, sensitivity, and F1-score. Results indicated that the Random Forest model outperformed the Decision Tree, achieving 83% accuracy in predicting preference categories. Feature importance analysis highlighted that price and physical dimensions were the most significant factors influencing consumer preferences. These findings suggest that practical and economic aspects are prioritized over aesthetic features when choosing furniture. The study demonstrates the potential of machine learning in predicting consumer behavior, offering valuable insights for manufacturers and retailers in optimizing product development, inventory management, and marketing strategies.
Download PDF
Full Article
Predicting Consumer Preferences for Furniture Products on E-commerce Platforms: An Analysis Using Machine Learning and Favorite Listing Data
Hüseyin Çardak,a Selahattin Bardak,b,* Timuçin Bardak,a Okan Capraz,a Sultan Ozcetin,a and Samet Kızılırmak c
The rapid growth of e-commerce platforms presents unique opportunities to analyze consumer behavior and predict product preferences in the furniture industry. This study explores the use of machine learning techniques to predict consumer choices for furniture products based on favorite listing data from e-commerce platforms. A dataset of 239 furniture products was collected, categorized into three groups: most preferred, moderately preferred, and least preferred. Key attributes, including furniture type, dimensions (width, depth, height), color, material, and price, were analyzed. Machine learning models, specifically Decision Trees and Random Forests, were applied to develop prediction models for these categories. The models were assessed using metrics such as accuracy, precision, sensitivity, and F1-score. Results indicated that the Random Forest model outperformed the Decision Tree, achieving 83% accuracy in predicting preference categories. Feature importance analysis highlighted that price and physical dimensions were the most significant factors influencing consumer preferences. These findings suggest that practical and economic aspects are prioritized over aesthetic features when choosing furniture. The study demonstrates the potential of machine learning in predicting consumer behavior, offering valuable insights for manufacturers and retailers in optimizing product development, inventory management, and marketing strategies.
DOI: 10.15376/biores.20.4.9768-9784
Keywords: Furniture industry; E-commerce; Data mining; Prediction
Contact information: a: Furniture and Decoration Program, Bartin Vocational School, Bartın University, 74200, Bartın, Turkey; b: Department of Computer Engineering, Faculty of Engineering and Architecture, Sinop University, 57000, Sinop, Turkey; c: Department of Forest Industrial Engineering, Faculty of Forestry, Bartin University, 74200, Bartın, Turkey; *Corresponding author: sbardak@sinop.edu.tr
INTRODUCTION
The growing influence of the Internet on modern life has led to a significant increase in online shopping. The large amounts of data collected on e-commerce platforms enable management to make informed decisions and gain deeper insights into consumer behavior. In addition, these forecasts enable more accurate financial budgeting, more effective operational planning, and more efficient inventory policies (Sun et al. 2008). Artificial intelligence (AI) has become an essential element of Industry 4.0 and is expected to maintain its importance with the advent of Industry 5.0. At the same time, companies are increasingly integrating AI technologies into all operational areas. This phenomenon reflects the profound impact of AI on industrial processes and business models (Sigov et al. 2022; de Waal et al. 2024). Machine learning (ML) algorithms have emerged as a powerful tool for solving complex industry-specific forecasting problems. However, research studies in the furniture industry are still limited. This study fills this gap by uniquely applying ML techniques to big data from e-commerce platforms, focusing on predicting consumer preferences in the furniture sector based explicitly on favorite listing behavior. This approach makes a unique contribution to the field, as it combines the power of ML with rich data obtained from online shopping platforms. Applying ML techniques makes it possible to analyze large amounts of data, identify patterns, and formulate predictions. The use of ML algorithms facilitates the acceleration of innovation by processing large data sets and running simulations (Xue et al. 2024).
Random forest (RT) is a widely used supervised ML algorithm. This algorithm can be effectively applied to classification and regression problems. Random forest has gained popularity among researchers and practitioners due to its adaptability to different data sets and high accuracy (Breiman 2001; Lee et al. 2024). Decision tree (DT) is a nonparametric supervised learning technique for classification or regression. It simplifies and visualizes complex decision processes by modeling patterns in the data set in a hierarchical structure. The DT works by repeatedly partitioning the data according to specific characteristics and making the most appropriate distinction at each node. This way, effective results can be achieved in classification and regression problems (Song and Ying 2015; Alakbari et al. 2023).
Over the past decade, e-commerce platforms such as Amazon and Alibaba have become indispensable tools for consumers searching for and purchasing products. The success of these platforms can be attributed to their sophisticated ability to collect and analyze consumer behavior data. E-commerce sites use recommendation systems to provide personalized product recommendations to individual users, thereby encouraging consumers to discover and purchase products that match their established shopping habits. This approach has the dual benefit of improving the user experience and increasing sales (Zhang et al. 2020; Zhou et al. 2024). In recent years, there has been a significant increase in research into using ML techniques to predict consumer behavior and future buying trends. These approaches use a variety of algorithms, such as logistic regression, decision trees, artificial neural networks, and support vector machines, to identify patterns in consumer data. The identified patterns are then used to create models that predict consumer preferences (Moro et al. 2014; Liu et al. 2024). To illustrate, the deep learning approach proposed by Gabel and Timoshenko (2022) extracts preference representations based on customers’ purchase history and uses this information to predict future product choices.
The furniture industry has complex variables that must be considered when predicting consumer preferences and behavior. Consumers’ furniture choices are influenced not only by the tangible characteristics of the products in question (e.g., dimensions, composition, color) but also by individual aesthetic tendencies and economic conditions. This requires using more sophisticated and accurate analytical tools to accurately predict the number of times furniture products will be added to the favorites list. Applying ML algorithms to extract meaningful insights from large data sets is a promising way to improve such predictions’ accuracy by overcoming traditional methods’ limitations.
In this study, the aim was to predict consumers’ preferences for furniture products using data from e-commerce platforms and ML algorithms. Furniture products were classified into three groups based on the number of times they had been added to the favorites list: highly preferred, moderately preferred, and low preferred. This approach is a unique contribution to the field, combining the power of ML with rich data from online shopping platforms.
The study used two major ML algorithms: Decision Trees and Random Forests. These algorithms were chosen based on their proven effectiveness in handling complex, multidimensional data and their ability to provide interpretable results. The study used a comprehensive dataset of 239 furniture products, including variables such as furniture type, dimensions (width, depth, height), color, primary material, and price. Predictive models were built and evaluated using standard classification metrics such as accuracy, precision, sensitivity, and F1 score. In addition, feature importance analysis was performed to determine the relative impact of different factors on consumer preferences. The results of this research have important practical implications for the furniture industry and could transform production planning, inventory management, and marketing strategies. By accurately predicting consumer preferences based on the number of times they are added to a favorites list, companies can optimize their operations, reduce waste, and better meet customer demands.
EXPERIMENTAL
Methodology
Data collection and pre-processing
This study utilized a comprehensive dataset of 239 furniture products, meticulously collected from several publicly accessible Turkish e-commerce platforms. Data included furniture type, dimensions, color, material, price, and preference category counts. The data were collected between December 11, 2022, and August 11, 2023.
The collected data underwent standard cleaning, missing value handling, and outlier correction. Table 1 shows Sample Furniture Product Data (excerpt).
Table 1. Sample Furniture Product Data (excerpt)
Data Categorization Using K-means Clustering
The K-means clustering algorithm was used to more objectively categorize furniture items based on the number of times they were added to the favorites list. K- means clustering was chosen because of its effectiveness in identifying natural groups in the data. Furniture items were categorized into three groups based on the number of times they were added to the favorites list: most preferred, moderately preferred, and least preferred. To address the recommendation for specific classification criteria, the K-means algorithm identified the following ranges for each category based on the ‘number of times added to favorites list’ attribute:
Least Preferred: 0 – 15231 favorite listings
Moderately Preferred: 15232 – 33193 favorite listings
Most Preferred: 33194 – 60568 favorite listings
These ranges were determined by the clustering process, which aimed to minimize the within-cluster variance and maximize the between-cluster variance, thereby creating distinct and meaningful groups. The ‘k’ parameter for the K-means algorithm was set to 3, as the aim was to categorize the furniture into three preference levels as per the study’s objective. This categorization was used as the target variable for the prediction models used in this study.
Model Building
In this study, consumer preferences for furniture products operated as the frequency with which items are added to users’ favorites—using Decision Trees (DT) and Random Forest (RF). These algorithms were selected due to their ability to handle mixed-type, multidimensional data and capture non-linear relationships while retaining interpretability for industry stakeholders. DT provides transparent, rule-based structures that clarify how attributes drive classification, whereas RF—an ensemble that aggregates the predictions of multiple decision trees—improves accuracy and mitigates overfitting relative to a single tree, yielding robust generalization on complex e-commerce data (Breiman 2001). Prior work further supports the use of DT/RF in e-commerce analytics and preference prediction (Haque 2024; Mustakim et al. 2024). Both models were trained on the collected e-commerce dataset containing categorical (e.g., type, color, primary material) and numerical (e.g., width, depth, height, price) features; the evaluation procedure and performance metrics are detailed in the following subsection. The dataset was divided into training and test sets, with 70% allocated for training and the remaining 30% for testing. This 70/30 split is a widely adopted standard in ML, balancing the need for sufficient training data to build robust models and enough testing data to evaluate model performance effectively (Pham et al. 2019). Such a division helps in mitigating overfitting and provides a reliable assessment of the model’s generalization capabilities.
Fig. 1. RapidMiner process (software interface screenshot) used to train and evaluate the models
In the context of furniture preference modeling, this approach aligns with methodologies employed in recent studies, such as Yu et al. (2023), who utilized a similar data partitioning strategy to analyze the relationship between consumer personality traits and preferences for wood furniture product characteristics. The models in this study were trained and evaluated using RapidMiner, a software platform widely recognized for its applicability in scientific research and ML tasks (Gonçalves et al. 2013; Mozaffarinya et al. 2019; Gonçalves et al. 2020; Sher et al. 2022). The process created for applying the models in the RapidMiner program is given in Fig. 1. RapidMiner process (software interface screenshot) was used to train and evaluate the models.
To optimize model performance, comprehensive hyperparameter tuning was conducted using grid search methodology through RapidMiner’s Optimize Parameters (Grid) operator. This operator systematically executed the subprocess for all combinations of selected parameter values to identify the optimal configuration that maximized prediction accuracy. Table 2 presents the three most critical hyperparameters for both Random Forest and Decision Tree models, along with their optimal values determined through this grid search process.
Table 2. Most Important Hyperparameters for Random Forest and Decision Tree Models
Model Evaluation
The model’s performance was comprehensively evaluated using accuracy, precision, and sensitivity, which are standard metrics for classification tasks in machine learning, providing a robust assessment of overall correctness, positive prediction reliability, and true positive identification, respectively (Sokolova and Lapalme 2009; Puri et al. 2017; Siering et al. 2018; Szabó et al. 2024).
Accuracy was calculated by dividing the total number of correctly classified observations by the total number of observations:
“Accuracy” = (“TP” + “TN”) / (“TP” + “FP” + “TN” + “FN”) (1)
Precision is a metric that measures the accuracy of a classifier, i.e., whether a sample classified as belonging to a certain class belongs to that class:
“Precision” = “TP” / (“TP” + “FP”) (2)
Sensitivity is defined as the ratio of true positive predictions to the number of positive samples:
“Sensitivity” = “TP” / (“TP” + “FN”) (3)
In Eq. 3, TP is True Positive; TN means True Negative; FP is False Positive; and FN denotes False Negative.
These metrics were used to compare the performances of different algorithms and permit them to select the best-performing model.
Feature Importance Analysis
Following the identification of the Random Forest model as the most robust predictor, a comprehensive feature importance analysis was conducted to quantify the influence of each attribute on consumer preferences for furniture products. This analysis was performed using the ‘Attribute Weights’ functionality inherent to the Random Forest operator in RapidMiner. The calculation of these weights is based on the principle of measuring the total reduction in node impurity (specifically, the Gini impurity criterion) that an attribute provides across all decision trees within the ensemble. For each attribute, its importance score is computed as the sum of the Gini impurity decreases for every node where that attribute was utilized for splitting the data. Consequently, attributes that are frequently selected for splitting and contribute significantly to the homogeneity of child nodes receive higher importance weights. This quantitative approach made it possible to objectively rank the factors influencing consumer choices, revealing the relative impact of attributes such as price, dimensions, and material on product favoritism. This detailed methodology ensures that the findings regarding feature importance are transparent and can be independently verified or applied by other researchers interested in similar analytical approaches.
After selecting the best-performing model (Random Forest; accuracy = 83.10%), the next step was to compute the global feature importance in RapidMiner using the Weight by Tree Importance operator. This method returns an ExampleSet with attributes and normalized weights (sum = 1). Each weight equals the total decrease in node impurity contributed by that attribute across all splits in the forest, weighted by the number of samples at each node; improvements are computed with the same splitting criterion as the model (Gini for classification). Only the following features were evaluated: Price, Width, Height, Depth, Color, Primary Material, and Furniture Type.
RESULTS AND DISCUSSION
This section reports the empirical findings and their implications. Using Random Forest (RF) and Decision Tree (DT), there was an evaluation of the predictive performance with accuracy, precision, sensitivity (recall), and F1-score. As summarized in Table 2, RF consistently outperformed DT: 83.10% overall accuracy for RF versus 73.24% for DT. Class-level F1-scores likewise favored RF (Class 0: 71.43%; Class 1: 70.59%; Class 2: 89.36%) over DT (Class 0: 53.33%; Class 1: 48.28%; Class 2: 83.67%), indicating better generalization across “most,” “moderately,” and “least” preferred categories. These gains are consistent with RF’s ensemble learning, which reduces overfitting relative to a single tree and captures non-linear interactions in mixed-type e-commerce data.
Following model assessment, a feature-importance analysis with the best-performing RF model showed that Price (0.304) and dimensional attributes—Width (0.228), Height (0.188), Depth (0.146)—were the dominant drivers of preference, whereas Color (0.073), Primary Material (0.036), and Furniture Type (0.025) were comparatively less influential. Practically, this implies that consumers prioritize economic and spatial constraints over aesthetic or categorical attributes when selecting furniture online; pricing and size-fit thus emerge as primary levers for product, inventory, and merchandising decisions.
Model Performance Comparison
Table 3 presents a detailed comparison of the performance metrics for both the RF and DT models when applied to the e-commerce dataset.
Table 3. RF vs. DT Performance on the E-commerce Dataset
The Random Forest Confusion Matrix (Table 4) and Decision Tree Confusion Matrix (Table 5) demonstrate the detailed performance breakdown.
Table 4. Random Forest Confusion Matrix
Table 5. Decision Tree Confusion Matrix
The RF model demonstrated superior overall performance with an accuracy rate of 83.10%, surpassing the DT model’s accuracy of 73.24%. This suggests that the RF algorithm more effectively captures complex patterns in e-commerce furniture preference data, consistent with recent findings showing the superior predictive capabilities of RF over DT (Helmud et al. 2024). In terms of model strengths and weaknesses, the RF model exhibited consistent performance across all preference categories and demonstrated high accuracy, particularly in identifying the least preferred furniture products. This consistency is attributed to RF’s ensemble learning approach, which mitigates overfitting and enhances generalization by aggregating predictions from multiple decision trees (Kinasih et al. 2025).
Conversely, the DT model showed solid performance in determining the least preferred products, while exhibiting lower overall accuracy. However, it encountered difficulties in distinguishing between the most and moderately preferred products.
The RF algorithm implemented in this study showcased remarkable success, meeting the performance criteria widely accepted in the literature (Sokolova and Lapalme 2009; Zhu et al. 2010; Luo et al. 2016). Its ability to handle high-dimensional data and provide robust predictions aligns with findings from other recent studies where RF models outperformed even more complex algorithms such as deep learning models in specific contexts, such as predicting furniture prices (Bardak 2023).
Feature Importance
In this study, the most powerful RF algorithm was used to determine the importance levels of the features that influence the number of favorites of furniture products. The factors analyzed included different characteristics such as height, width, depth, color, primary material, price, and type of furniture. Table 6 shows in detail the important weights of these factors obtained from the e-commerce platforms data using the RF algorithm. These results contribute significantly to a more comprehensive understanding of the factors that shape consumer preferences in the furniture sector and can provide valuable insights to stakeholders in the sector.
Table 6. Weights of Factors Based on Furniture Data Obtained through the Random Forest Algorithm
The RF algorithm’s analysis of furniture attributes reveals significant insights into consumer preferences in the furniture industry. Price emerges as the dominant factor with the highest importance weight (0.304), indicating strong consumer price sensitivity. This is closely followed by dimensional attributes – width, height, and depth – suggesting that the physical size of furniture plays a critical role in purchase decisions, likely due to space constraints or aesthetic considerations. Color is moderately important, while furniture material and type are relatively unimportant. This hierarchy of factors suggests that consumers prioritize practical and economic aspects over aesthetic or categorical features when selecting furniture.
This finding aligns with previous research, where price was found to be a decisive factor in furniture purchasing decisions and RF models demonstrated high accuracy in predicting furniture prices (Bardak 2023). Dimensional attributes particularly width, height, and depth also significantly impact consumer decisions. Gudarzi et al. (2022) similarly found that physical dimensions of furniture are critical considerations for consumers, often outweighing other product characteristics in importance.
While color holds moderate importance, material and type appear to be less influential factors. Supporting this, Guzel (2020) reported that consumers in Kayseri, Turkey, frequently prioritize affordable and functional composite furniture over more expensive solid wood alternatives. Moreover, Yu et al. (2023) highlighted that consumer personality traits such as extraversion and conscientiousness significantly influence the perceived importance of product features like quality and design, although these factors still ranked below economic and dimensional considerations.
This hierarchy of factors suggests that for the furniture products analyzed in this study (coffee tables and TV units), practical and economic aspects such as price and physical dimensions are significant factors influencing consumer preferences. While these findings highlight the importance of practical and economic factors within our specific dataset, it is important to acknowledge that consumer preferences for furniture can be highly nuanced and influenced by various other factors, including aesthetic considerations, specific furniture categories (e.g., functional vs. decorative), and diverse user demographics (e.g., young people, parents). Future research could explore these aspects in more detail by examining a broader range of furniture categories and demographic segments.
CONCLUSIONS
The aim of this study was to contribute to the existing literature on the application of machine learning in consumer behavior analysis, with a particular focus on the furniture industry.
- This study examined the potential of machine learning algorithms in predicting consumer preferences for furniture products using e-commerce favoriting data. The research aimed to estimate the number of favorites of furniture products using the capabilities of Decision Trees and Random Forest algorithms. The analysis demonstrated that the RF algorithm exhibited superior performance (accuracy of 83.10%) compared to the DT model (73.24% accuracy).
- Feature importance analysis using the RF model revealed that price and the physical dimensions of the furniture (width, height and depth) significantly influenced consumer preferences. These findings highlight the importance of practical considerations, such as space and financial constraints, in e-commerce furniture preference decisions. In contrast, factors such as furniture type and material were found to have a relatively limited influence on consumer choice. This suggests that aesthetic or categorical features may be secondary to functional attributes.
- These findings may provide valuable insights for furniture manufacturers and retailers, enabling them to optimize product development, inventory management, and marketing strategies based on predictive analysis.
- The successful application of machine learning techniques in this study demonstrates their potential to contribute to decision-making processes in the furniture industry. As e-commerce platforms continue to generate large amounts of consumer data, the integration of advanced analytics will be critical to driving business success and responding to changing market trends.
It should be noted that the present findings are specific to the functional furniture categories examined (coffee tables and TV units) and may not be generalizable to decorative furniture items or different consumer demographic groups. Future studies should investigate how preferences vary across different furniture categories and user segments to provide a more comprehensive understanding of consumer behavior in the furniture industry.
ACKNOWLEDGMENTS
This study was supported by the Scientific and Technological Research Council (TUBITAK) of Turkey (2209-A BIDEB, Funding Number: 1919B012202575).
REFERENCES CITED
Alakbari, F. S., Mohyaldinn, M. E., Ayoub, M. A., Salih, A. A., and Abbas, A. H. (2023). “A decision tree model for accurate prediction of sand erosion in elbow geometry,” Heliyon 9(7), article 17639. DOI: 10.1016/j.heliyon.2023.e17639
Bardak, T. (2023). “Predicting prices of case furniture products using web mining techniques,” BioResources 18(4), 7412-7427. DOI: 10.15376/biores.18.4.7412-7427
Breiman, L. (2001). “Random forests,” Machine Learning 45(1), 5-32. DOI: 10.1023/A:1010933404324arno.uvt.nl
de Waal, H., Nyawa, S., and Wamba, S. F. (2024). “Consumers’ financial distress: Prediction and prescription using interpretable machine learning,” Information Systems Frontiers 1-22. DOI: 10.1007/s10796-024-10501-1
Gabel, S., and Timoshenko, A. (2022). “Product choice with large assortments: A scalable deep-learning model,” Management Science 68(3), 1808-1827. DOI: 10.1287/mnsc.2021.3969
Gonçalves, C., Ferreira, D., Neto, C., Abelha, A., and Machado, J. (2020). “Prediction of mental illness associated with unemployment using data mining,” Procedia Computer Science 177, 556-561. DOI: 10.1016/j.procs.2020.10.078
Gonçalves, M., Morreale, P., and Silva, C. (2013). “Analysis and visualization of large-scale time series network data,” in: Modeling and Processing for Next-Generation Big-Data Technologies, pp. 211-229. DOI: 10.1007/978-3-319-09177-89
Gudarzi, E., Tajdini, A., Pourmousa, S., Latibari, A. J., and Roohnia, M. (2022). “Structural equation modeling (SEM) of consumer preferences for wooden furniture (home and office),” BioResources 17(1), 1551-1565. DOI: 10.15376/biores.17.1.1551-1565
Guzel, T. A. (2020). “Consumer attitudes toward preference and use of wood, woodenware, and furniture: A sample from Kayseri, Turkey,” BioResources 15(1), 28-37. DOI: 10.15376/biores.15.1.28-37
Haque, M. Z. (2024). “E-commerce product recommendation system based on ML algorithms,” https://arxiv.org/abs/2407.21026 accessed 28 April 2025.
Helmud, E., Fitriyani, F., and Romadiana, P. (2024). “Classification comparison performance of supervised machine learning random forest and decision tree algorithms using confusion matrix,” Jurnal SISFOKOM 13(1), 92-97. DOI: 10.32764/jsfk.v13i1.4291
Kinasih, A. P., Sulistyo, S., and Arifianto, A. (2025). “Comparative analysis of decision tree and random forest classifiers, “ Science in Information Technology Letters 5(2), 13-24. DOI: 10.31763/sitech.v5i2.1746
Lee, J., Kim, J., Hahn, S., Han, H., Shin, G., Kim, W.-C., and Yoon, S.-W. (2024). “Data-driven disruption prediction using random forest in KSTAR,” Fusion Engineering and Design 199, article 114128. DOI: 10.1016/j.fusengdes.2023.114128
Liu, Y., Li, B., Yang, S., and Li, Z. (2024). “Handling missing values and imbalanced classes in machine learning to predict consumer preference: Demonstrations and comparisons to prominent methods,” Expert Systems with Applications 237, article 121694. DOI: 10.1016/j.eswa.2023.121694
Luo, J., Wu, M., Gopukumar, D., and Zhao, Y. (2016). “Big data application in biomedical research and health care: a literature review,” Biomedical Informatics Insights 8, BII-S31559. DOI: 10.4137/BII.S31559
Moro, S., Cortez, P., and Rita, P. (2014). “A data-driven approach to predict the success of bank telemarketing,” Decision Support Systems 62, 22-31. DOI: 10.1016/j.dss.2014.03.001
Mozaffarinya, M., Reza Shahriyari, A., Karim Bahadori, M., Ghazvini, A., Shamsadin Athari, S., and Vahedi, G. (2019). “A data-mining algorithm to assess key factors in asthma diagnosis,” Revue Française d’Allergologie 59(7), 487-492. DOI: 10.1016/j.reval.2019.01.013
Mustakim, N. A., Abdul Aziz, M., and Abdul Rahman, S. (2024). “Predicting consumer behavior in e-commerce using decision tree: A case study in Malaysia,” Information Management and Business Review 16(3), 201-209. DOI: 10.22610/imbr.v16i3(I).3965
Pham, B. T., Jaafari, A., Prakash, I., and Bui, D. T. (2019). “A novel hybrid intelligent model of support vector machines and the MultiBoost (MB-SVM) for landslide susceptibility modeling,” Bulletin of Engineering Geology and the Environment, 78(4), 2865-2886. DOI: 10.1007/s10064-018-1281-y
Puri, C., Akhil, Gaurav, Kush, and Kumar, N. (2017). “Opinion ensembling for improving economic growth through tourism,” Procedia Computer Science 122, 237-244. DOI: 10.1016/j.procs.2017.11.366
Sher, T., Rehman, A., and Kim, D. (2022). “COVID-19 outbreak prediction by using machine learning algorithms,” Computers, Materials and Continua 74(1), 1561-1574. DOI: 10.32604/cmc.2023.032020
Siering, M., Deokar, A. V, and Janze, C. (2018). “Disentangling consumer recommenda-tions: Explaining and predicting airline recommendations based on online reviews, “ Decision Support Systems 107, 52-63. DOI: 10.1016/j.dss.2018.01.002
Sigov, A., Ratkin, L., Ivanov, L. A., and Xu, L. Da. (2022). “Emerging enabling technologies for industry 4.0 and beyond,” Information Systems Frontiers 1–11. DOI: 10.1007/s10796-021-10213-w
Sokolova, M., and Lapalme, G. (2009). “A systematic analysis of performance measures for classification tasks,” Information Processing & Management 45(4), 427-437. DOI: 10.1016/j.ipm.2009.03.002
Song, Y. Y., and Ying, L. U. (2015). “Decision tree methods: Applications for classification and prediction,” Shanghai Archives of Psychiatry 27(2), article 130. DOI: 10.11919/j.issn.1002-0829.215044
Sun, Z. L., Choi, T. M., Au, K. F., and Yu, Y. (2008). “Sales forecasting using extreme learning machine with applications in fashion retailing,” Decision Support Systems 46(1), 411-419. DOI: 10.1016/j.dss.2008.07.009
Szabó, S., Holb, I. J., Abriha-Molnár, V. É., Szatmári, G., Singh, S. K., and Abriha, D. (2024). “Classification Assessment Tool: A program to measure the uncertainty of classification models in terms of class-level metrics,” Applied Soft Computing 155, article 111468. DOI: 10.1016/j.asoc.2024.111468
Xue, J., Alinejad-Rokny, H., and Liang, K. (2024). “Navigating micro- and nano-motors/swimmers with machine learning: Challenges and future directions,” ChemPhysMater 3(3), 273-283. DOI: 10.1016/j.chphma.2024.06.001
Yu, S., Zheng, Q., Chen, T., Zhang, H., and Chen, X. (2023). “Consumer personality traits vs. their preferences for the characteristics of wood furniture products,” BioResources 18(4), 7443-7459. DOI: 10.15376/biores.18.4.7443-7459
Zhang, J., Adomavicius, G., Gupta, A., and Ketter, W. (2020). “Consumption and performance: Understanding longitudinal dynamics of recommender systems via an agent-based simulation framework,” Information Systems Research 31(1), 76-101.
Zhou, F., Jiang, Y., Qian, Y., Liu, Y., and Chai, Y. (2024). “Product consumptions meet reviews: Inferring consumer preferences by an explainable machine learning approach,” Decision Support Systems 177, article 114088. DOI: 10.1016/J.DSS.2023.114088
Zhu, W., Zeng, N., and Wang, N. (2010). “Sensitivity, specificity, accuracy, associated confidence interval and ROC analysis with practical SAS implementations,” NESUG Proceedings: Health Care and Life Sciences, Baltimore, Maryland 19, 67.
Article submitted: May 8, 2025; Peer review completed: August 13, 2025; Revised version received: September 5, 2025; Accepted: September 16, 2025; Published: September 23, 2025.
DOI: 10.15376/biores.20.4.9768-9784
APPENDIX: Data Table
