Binary Classifier Calibration using an Ensemble of Near Isotonic Regression Models

Calibrating a Transformer-Based Model's Confidence on Community-Engaged Research Studies: Decision Support Evaluation Study

BACKGROUND

Deep learning offers great benefits in classification tasks such as medical imaging diagnostics or stock trading, especially when compared with human-level performances, and can be a viable option for classifying distinct levels within community-engaged research (CEnR). CEnR is a collaborative approach between academics and community partners with the aim of conducting research that is relevant to community needs while incorporating diverse forms of expertise. In the field of deep learning and artificial intelligence (AI), training multiple models to obtain the highest validation accuracy is common practice; however, it can overfit toward that specific data set and not generalize well to a real-world population, which creates issues of bias and potentially dangerous algorithmic decisions. Consequently, if we plan on automating human decision-making, there is a need for creating techniques and exhaustive evaluative processes for these powerful unexplainable models to ensure that we do not incorporate and blindly trust poor AI models to make real-world decisions.

OBJECTIVE

We aimed to conduct an evaluation study to see whether our most accurate transformer-based models derived from previous studies could emulate our own classification spectrum for tracking CEnR studies as well as whether the use of calibrated confidence scores was meaningful.

METHODS

We compared the results from 3 domain experts, who classified a sample of 45 studies derived from our university's institutional review board database, with those from 3 previously trained transformer-based models, as well as investigated whether calibrated confidence scores can be a viable technique for using AI in a support role for complex decision-making systems.

RESULTS

Our findings reveal that certain models exhibit an overestimation of their performance through high confidence scores, despite not achieving the highest validation accuracy.

CONCLUSIONS

Future studies should be conducted with larger sample sizes to generalize the results more effectively. Although our study addresses the concerns of bias and overfitting in deep learning models, there is a need to further explore methods that allow domain experts to trust our models more. The use of a calibrated confidence score can be a misleading metric when determining our AI model's level of competency.

An Explainable Machine Learning Model for Material Backorder Prediction in Inventory Management

Global competition among businesses imposes a more effective and low-cost supply chain allowing firms to provide products at a desired quality, quantity, and time, with lower production costs. The latter include holding cost, ordering cost, and backorder cost. Backorder occurs when a product is temporarily unavailable or out of stock and the customer places an order for future production and shipment. Therefore, stock unavailability and prolonged delays in product delivery will lead to additional production costs and unsatisfied customers, respectively. Thus, it is of high importance to develop models that will effectively predict the backorder rate in an inventory system with the aim of improving the effectiveness of the supply chain and, consequentially, the performance of the company. However, traditional approaches in the literature are based on stochastic approximation, without incorporating information from historical data. To this end, machine learning models should be employed for extracting knowledge of large historical data to develop predictive models. Therefore, to cover this need, in this study, the backorder prediction problem was addressed. Specifically, various machine learning models were compared for solving the binary classification problem of backorder prediction, followed by model calibration and a post-hoc explainability based on the SHAP model to identify and interpret the most important features that contribute to material backorder. The results showed that the RF, XGB, LGBM, and BB models reached an AUC score of 0.95, while the best-performing model was the LGBM model after calibration with the Isotonic Regression method. The explainability analysis showed that the inventory stock of a product, the volume of products that can be delivered, the imminent demand (sales), and the accurate prediction of the future demand can significantly contribute to the correct prediction of backorders.

Calibrating Classification Probabilities with Shape-Restricted Polynomial Regression

In many real-world classification problems, accurate prediction of membership probabilities is critical for further decision making. The probability calibration problem studies how to map scores obtained from one classification algorithm to membership probabilities. The requirement of non-decreasingness for this mapping involves an infinite number of inequality constraints, which makes its estimation computationally intractable. For the sake of this difficulty, existing methods failed to achieve four desiderata of probability calibration: universal flexibility, non-decreasingness, continuousness and computational tractability. This paper proposes a method with shape-restricted polynomial regression, which satisfies all four desiderata. In the method, the calibrating function is approximated with monotone polynomials, and the continuously-constrained requirement of monotonicity is equivalent to some semidefinite constraints. Thus, the calibration problem can be solved with tractable semidefinite programs. This estimator is both strongly and weakly universally consistent under a trivial condition. Experimental results on both artificial and real data sets clearly show that the method can greatly improve calibrating performance in terms of reliability-curve related measures.

Binary Classifier Calibration Using an Ensemble of Piecewise Linear Regression Models

In this paper we present a new non-parametric calibration method called ensemble of near isotonic regression (ENIR). The method can be considered as an extension of BBQ (Pakdaman Naeini, Cooper and Hauskrecht, 2015b), a recently proposed calibration method, as well as the commonly used calibration method based on isotonic regression (IsoRegC) (Zadrozny and Elkan, 2002). ENIR is designed to address the key limitation of IsoRegC which is the monotonicity assumption of the predictions. Similar to BBQ, the method post-processes the output of a binary classifier to obtain calibrated probabilities. Thus it can be used with many existing classification models to generate accurate probabilistic predictions. We demonstrate the performance of ENIR on synthetic and real datasets for commonly applied binary classification models. Experimental results show that the method outperforms several common binary classifier calibration methods. In particular, on the real data we evaluated, ENIR commonly performs statistically significantly better than the other methods, and never worse. It is able to improve the calibration power of classifiers, while retaining their discrimination power. The method is also computationally tractable for large scale datasets, as it is O(N logN) time, where N is the number of samples.