Abstract
One of the importance of the contamination uncertainty model is to consider in-determinism in the uncertainty. We consider this advanced property and develop two methods. These methods identify if there is imprecision in a given model or data. In the first approach, we build two different—a probability distribution and an interval—models for a test function $f$ via given data/model. Then, we identify the level of imprecision by assessing, so-called model trust, $\epsilon \in (0,1)$ in the contamination model whether the weight is higher for the probabilistic/interval model or not. In the second approach, we calculate the lowest and highest previsions for the test function and identify the imprecision interval out of them. We further discuss and show the idea via two simple production and clutch design problems to illustrate our novel results.
Introduction
In classical probability theory, an agent is typically assumed to be "decisive"—given sufficient data, one can calculate a precise probability for any event. However, real-world engineering data is frequently vague, incomplete, or influenced by unpredictable factors such as human behavior or environmental fluctuations. Traditional statistical approaches often attempt to force a single "best guess" probability onto this uncertain data, an approach that can be dangerously unrealistic.
Researchers Keivan Shariatmadar, Hans Hallez, and David Moens argue for a fundamental distinction between risk (where probabilities are known) and imprecision (where the probabilities themselves are uncertain). Most standard models conflate these, leading to "optimal" solutions that fail under slight real-world shifts. This paper addresses the challenge of mathematically identifying and measuring the level of imprecision within a dataset.
The $\epsilon$-Contamination Model
The core framework utilized is the $\epsilon$-contamination model. This model functions as a weighted average between a precise probabilistic model and an imprecise interval model:
- The Probabilistic Model ($E$): A standard distribution representing the best statistical estimate.
- The Interval Model ($\underline{E}$): A range of possible values representing "vacuous" or worst-case uncertainty.
The parameter $\epsilon$ (ranging from 0 to 1) represents "model trust" or the weight assigned to imprecision.
Methodological Approaches
The authors propose two distinct methods to quantify $\epsilon$:
- Method I (Direct Identification): If a known outcome for a test function exists, $\epsilon$ can be solved for directly. An $\epsilon > 0.5$ indicates that imprecision is the dominant factor in the data.
- Method II (Prevision Intervals): This method calculates "lower and upper previsions"—the highest price one would pay to buy a gamble (uncertain outcome) and the lowest price one would accept to sell it. The gap between these values defines the "imprecision interval."
Applications and Results
The validity of these methods was tested across several engineering and logistics scenarios.
Logistics and Traffic
Using factory travel duration data, the study demonstrated that a single probability curve fails to capture the "indeterminism" caused by varying weather or driver behavior. The $\epsilon$-contamination model allows for a "neighborhood" of probabilities, providing a more robust representation than purely Bayesian models.
Industrial Production and Mechanical Design
The methods were applied to two specific problems:
- Chocolate Production: In a synthetic linear programming problem regarding cacao availability, Method II revealed a near-certainty of imprecision ($\epsilon^* \approx 0.9999$). This suggests that relying on simple averages for resource constraints would lead to frequent production failures.
- Clutch Design: Analyzing the friction coefficient ($\mu$) in wet-plate clutches—a parameter notoriously difficult to stabilize—again showed a near 100% chance of imprecision. By identifying this, engineers can design for a range of friction values, ensuring safe operation despite fluctuations in oil temperature or centrifugal forces.
Conclusion
The $\epsilon$-contamination model provides a powerful "knob" for decision-makers, allowing them to tune how much they trust a probabilistic guess versus hedging against a wide interval of possibilities. By treating "indecision" as a feature rather than a failure, this approach yields more robust engineering designs. Future work will focus on applying these methods to highly non-linear, high-dimensional systems and refining the estimation of the initial probabilistic and interval components.
Summary updated