Statistical analysis of data collected at microscale levels often presents significant challenges when researchers attempt to derive inferences at a macroscale level while attempting to preserve as much original information as possible. A recently developed symbolic interval-valued regression model offers a novel solution to this problem, providing a robust statistical framework for constructing relationships between two or more interval-valued variables. By organizing micro-scale data into intervals using the largest and smallest order statistics, researchers can now maintain the natural ordering requirements—where a higher value of the dependent variable remains mathematically greater than its lower counterpart—without the need to impose artificial constraints that previously limited the efficacy of similar models, particularly when dealing with negative values.
Traditional methods, such as the initial linear regression models developed by Billard and Diday or the center and range method (CRM) proposed by Lima-Neto and De Carvalho, have faced recurring issues. Many of these established approaches could not guarantee that the higher value of the dependent variable would be mathematically greater than the lower value if the data included negative figures. While subsequent attempts, such as the constrained center and range method, sought to fix these issues by assuming all regression model parameters for range intervals were positive, these constraints were often restrictive. The new approach follows the conceptual framework of utilizing order statistics to reorganize micro-data, effectively ensuring that the natural order of the variables is preserved intrinsically through the model properties rather than through external, rigid parameters.
The mathematical foundation of this symbolic regression model involves the development of innovative maximum likelihood estimators (MLEs) for both simple linear and multiple linear interval-valued regression scenarios. The research demonstrates that these estimators exhibit asymptotic normality, a property that allows for rigorous statistical testing and confidence estimation. By deriving the Fisher information matrix, the researchers provide a method to assess the significance of the model parameters. Crucially, the model does not require the intervals to be uniformly distributed, as the authors demonstrate that the approach works well with normal distribution assumptions and can be extended to other distributions, providing greater flexibility than many legacy methods.
To validate the efficacy of the proposed method, the study conducted comprehensive Monte Carlo simulations under three distinct scenarios. In the first scenario, the dependent variables were derived from the sum of independent variable intervals and residuals. In the second, researchers generated raw random samples and obtained dependent variables before reorganization into intervals. A third, highly practical scenario involved asynchronous data collection, where dependent and independent variables were sampled at different intervals. Across all scenarios, the simulation results highlighted the superior precision of the proposed model compared to conventional central methods. For example, in the initial scenarios, root mean squared errors (RMSEs) for slope estimators remained consistently low, often less than ten percent, revealing a significant performance improvement over traditional ordinary least squares applications.
The performance metrics from these simulations suggest that the new model excels particularly in scenarios where data collection frequencies differ. In the third simulation scenario, where the researchers simulated asynchronous independent and dependent variables, the proposed method yielded highly accurate estimations, with RMSEs for the intercept near ten percent and slope estimations falling below four percent. In stark contrast, estimators based on standard average values in the same simulation were highly inaccurate, with errors exceeding fifty percent. This disparity underscores the value of the new approach for environmental or medical datasets where sensor collection rates may vary significantly, as it accounts for the variability of the intervals rather than merely relying on block averages.
To further demonstrate the utility of the approach, the research applied the model to a real-world cardiological dataset originally studied by Raju, Billard, and Diday, which included pulse rates and systolic and diastolic blood pressures for eleven patients. By using the interval-valued regression model to predict pulse rates based on blood pressure intervals, the researchers compared their results against the CM model, the MinMax model, and the CRM model. The leave-one-out cross-validation results showed that the proposed method produced lower and more consistent RMSE values for both the upper and lower bounds of the predictions compared to the existing models. This suggests that by capturing the interconnectedness between the maximum and minimum values of the intervals, the proposed method provides a more accurate reflection of the underlying physiological relationships.
A second real-world application focused on the relationship between PM2.5 concentrations and weather variables, including wind speed and temperature, collected in southern Taiwan. Because the PM2.5 data were collected hourly while the meteorological variables were measured every ten minutes, the data were inherently asynchronous. The research successfully reorganized these observations into interval-valued datasets using order statistics. The fitted models provided intuitive results, confirming that higher wind speeds were associated with lower PM2.5 concentrations, while higher temperatures tended to correlate with higher pollution levels. The RMSE values obtained through this model further confirmed its stability and suitability for handling such asynchronous, real-world data streams where missing or misaligned observations could otherwise compromise standard regression techniques.
These developments point to a broader shift in how researchers can extract meaningful statistical insights from micro-scale observations. By moving beyond the limitations of standard linear models and avoiding the pitfalls of rigid constraints, this symbolic interval-valued regression model provides a reliable, mathematically rigorous way to handle interval data. The study establishes that the model is well-suited for diverse applications, ranging from cardiological monitoring to environmental quality assessment. As the research notes, because data points like temperature are often temporally dependent, future iterations of this model could be integrated with existing interval-valued time series regression frameworks to analyze even more complex, dynamic data structures.
The study concludes that the derivation of likelihood functions and the verification of asymptotic normality for these estimators offer a firm foundation for future statistical research in symbolic data analysis. By effectively handling the inherent noise and asynchronous nature of modern sensor data, the proposed models offer researchers a pathway to gain higher resolution insights from aggregated datasets. While the current model has proven effective for linear relationships, the flexibility inherent in its maximum likelihood approach suggests potential for further expansion into non-linear symbolic interval regression, potentially broadening its reach into even more specialized technical and scientific fields where interval-valued data are a standard unit of measurement.
The model establishes asymptotic normality through the derivation of the Fisher information matrix, providing a theoretical backbone for significance testing. By deriving the likelihood function for parameters θ = (β0, β1, σ²), the study demonstrates that these maximum likelihood estimators provide reliable inferential statistics. The researchers specifically calculated the Fisher information matrix to assess the significance of the intercept and slope parameters, ensuring that the model parameters are both identifiable and statistically sound. A key contribution of this study is the use of order statistics to transform micro-scale observations into intervals. This transformation allows the model to leverage the underlying distribution of the residuals effectively. In the study, the researchers utilized the normal distribution to demonstrate these results, while noting that the model’s structural design facilitates extension to other non-uniform distribution types. This flexibility represents a significant departure from many legacy methods, which often required strict distribution assumptions or enforced positive parameter constraints that did not always align with the reality of negative data values or complex data distributions.
Simulation studies were essential for confirming that these theoretical derivations held up under varied data generation scenarios. In the first scenario, researchers synthesized dependent variables by aggregating independent variable intervals and residuals, confirming the validity of the proposed regression model. The subsequent scenario, which mirrored real-world conditions where the largest order statistics of dependent and independent variables may not coincide, required the application of an absolute value operation to the integral term to ensure a positive likelihood function. This nuance highlights the model’s adaptability in handling sampling discrepancies between variables. Results consistently showed that the proposed method produced root mean squared errors for slope estimators that were significantly lower than those achieved by conventional central methods. For example, while standard methods struggled with the asynchronous sampling inherent in the third scenario, the proposed model maintained RMSE values for beta parameters consistently low, often around ten percent or less. This stability is critical, as conventional ordinary least squares estimation on averaged block data yielded errors exceeding fifty percent, illustrating the model’s superior precision in preserving the relationships between the original micro-scale data points.
Beyond theoretical and simulation-based performance, the model demonstrated robustness in analyzing real-world health and environmental datasets. In the cardiological study, which focused on predicting pulse rates from systolic and diastolic blood pressure, the proposed model’s ability to capture the interconnectedness between interval endpoints proved advantageous. While existing methods have been standard in the field, the new model provided more consistent root mean square error values for both upper and lower bounds. Similarly, the study’s analysis of PM2.5 concentrations in southern Taiwan showcased the model’s utility for asynchronous environmental data. By reorganizing hourly concentration data and ten-minute meteorological measurements into coherent intervals, the researchers identified clear associations between wind speed, temperature, and pollution levels. These empirical findings confirmed that the model not only performs well in controlled simulations but also offers a reliable, flexible approach to analyzing complex, irregularly sampled data streams common in modern sensor-driven research. The ability to model these relationships without losing the nuance contained in micro-level intervals marks a practical step forward for symbolic data analysis.
Subject of Research: Statistical modeling of symbolic interval-valued data.
Article Title: Symbolic Interval-Valued Regression Model
Article References: Lin, L.-C. (2026). Symbolic Interval-Valued Regression Model. Journal of Statistical Theory and Applications, 25(1), Article 26. https://doi.org/10.1007/s44199-026-00177-8
Image Credits: AI Generated
DOI: 10.1007/s44199-026-00177-8
Keywords: symbolic data analysis, interval-valued regression, maximum likelihood estimation, Fisher information matrix, asymptotic normality, statistical modeling, microscale data, macroscale data, order statistics, cardiological data, PM2.5 concentration, asynchronous data, linear regression, regression models, statistical inference, data science, variable correlation, interval data analysis
Cite Scienmag News
Blake Davidson. (November 18, 2026). New Symbolic Regression Model Effectively Analyzes Complex Interval-Valued Datasets. Scienmag. https://scienmag.com/new-symbolic-regression-model-effectively-analyzes-complex-interval-valued-datasets/
Blake Davidson. "New Symbolic Regression Model Effectively Analyzes Complex Interval-Valued Datasets." Scienmag, 18 November 2026, https://scienmag.com/new-symbolic-regression-model-effectively-analyzes-complex-interval-valued-datasets/. Accessed 18 November 2026.
Blake Davidson. "New Symbolic Regression Model Effectively Analyzes Complex Interval-Valued Datasets." Scienmag. November 18, 2026. https://scienmag.com/new-symbolic-regression-model-effectively-analyzes-complex-interval-valued-datasets/








