Symbolic Data Analysis and Interval-Valued Forecasting

Symbolic Data Analysis (SDA) extends classical statistical methods to handle complex data units that are not single values but symbols representing distributions, intervals, lists or histograms. By treating each symbol as the basic analytical unit, SDA enables researchers to capture within-group variability, relationships among distributional summaries and multiscale phenomena. Interval-valued forecasting is a specialised branch of time series analysis in which each observation is an interval rather than a point estimate, thereby reflecting measurement imprecision, aggregation effects or intrinsic uncertainty. Methods for interval forecasting draw on set-valued statistical models, singular spectrum analysis, support vector regression and neural networks to produce lower and upper bounds that encapsulate possible future trajectories. This dual focus on distributional data structures and interval predictions has found applications in finance, environmental monitoring, public health and engineering, where risk assessment and decision-making require transparent quantification of uncertainty. Advances in algorithmic efficiency, likelihood-based inference for aggregated data and ensemble forecasting techniques are broadening the scope of SDA and interval forecasting to large-scale and real-time contexts, reinforcing their global relevance for reliable, interpretable analytics.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Recent work has introduced generalised likelihood constructions for symbolic data, allowing inference on underlying measurement-level variables when only aggregated distributional summaries are observed. This framework supports novel symbol designs and unifies classical and symbolic analyses, facilitating scalable modelling of very large and complex datasets.

A computationally feasible neural network methodology has been developed to perform imprecise regression with interval-valued dependent variables. By iteratively optimising parameters of single-layer and multi-layer interval neural networks, the approach yields rigorous lower and upper predictive bounds, directly modelling epistemic uncertainty in outputs.

An extension of singular spectrum analysis for interval-valued time series has been proposed, decomposing interval observations into trend, cyclical and noise components. Forecasts are generated via linear recurrent relations, with demonstrated performance in tracking and predicting financial market dynamics under turbulent conditions.

Symbolic Data Analysis and Interval-Valued Forecasting publication trend

The graph below shows the total number of articles in symbolic data analysis and interval-valued forecasting across all publications each year (not limited to Nature Index journals).

Technical terms

Symbolic data analysis: A statistical paradigm in which data units are symbols that encapsulate distributions, intervals or complex summaries rather than single values, allowing inference on aggregated and multiscale phenomena.

Interval-valued data: Observations expressed as intervals [lower bound, upper bound] to represent measurement imprecision, aggregation effects or inherent uncertainty in the recorded values.

Interval forecasting: The process of predicting future values as intervals, producing lower and upper bounds that account for uncertainty and variability in time series or other sequential data.

Epistemic uncertainty: Uncertainty arising from lack of knowledge or imprecision in measurements, which can be represented and propagated through interval or set-valued models.

Singular spectrum analysis: A non-parametric technique for decomposing time series into interpretable components (trend, cycle, noise) by embedding the series in a trajectory matrix and performing singular value decomposition.

References

1. New models for symbolic data analysis. Advances in Data Analysis and Classification (2022).
2. Neural network model for imprecise regression with interval dependent variables. Neural Networks (2023).
3. Modeling interval trendlines: Symbolic singular spectrum analysis for interval time series. Journal of Forecasting (2021).
4. Support Vector Regression with Interval-Input Interval-Output. International Journal of Computational Intelligence Systems (2008).