Is My Model Up-to-date - Detecting CoViD-19 Variants by Machine Learning - Online drift detection and model update

Aug 3, 2026

Upload by Hazel

Machine learning systems can become outdated when the real-world data distribution shifts after deployment. Continuous adaptation is needed to preserve prediction quality. This work evaluates the behavior of the currently deployed model in an online fashion using a drift detector to capture performance slumps and trigger replacement with an up-to-date model. Experiments use 8642 hematochemical examinations from hospitalized patients and predict RT-PCR CoViD-19 outcomes, achieving an AUC of 0.794 with gains over offline and standard online binary approaches.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages8

Reading Time20 min

AccessFree to download

Abstract

Introduction

Problem of model obsolescence and drift

Need for detection and timely retraining

Related Work

Background on ML training/validation and performance degradation

Online learning, MLOps, and adaptive frameworks

Proposed Approach

Online evaluation with drift detection

Implementation and Deployment

Prototype details

On-the-field deployment

Conclusions

Is My Model Up-to-date? Detecting CoViD-19 Variants by Machine Learning⋆ Oguzhan Avci1,†, Giuseppe Pozzi1, **,† 1 DEIB, Politecnico di Milano, P. za L. da Vinci 32, I-20133, Milano, Italy Abstract Machine learning extracts models from huge quantities of data. Models trained and validated over past data can be deployed in making forecasts as well as in classifying new incoming data. The real world which generates data may change over time, making the deployed model an obsolete one. To preserve the quality of the currently deployed model, continuous machine learning is required. Our approach retrospectively evaluates in an online fashion the behaviour of the currently deployed model. A drift detector detects any performance slump, and, in case, can replace the previous model with an up-to-date one. The approach experiments on a dataset of 8642 hematochemical examinations from hospitalized patients gathered over 6 months: the outcome of the model predicts the RT-PCR test result about CoViD-19 . The method reached an area under the curve (AUC) of 0.794 , 6% better than offline and 5% better than standard online-binary classification techniques. Keywords Machine learning, adaptive modelling, concept drift, model update, CoViD-19 1. Introduction Artificial Intelligence (AI) initially aimed at simulating and replicating the human way of thinking.AI evolved very rapidly, and nowadays aims to “reason about huge quantities of data” [1], rising the concept of Machine Learning (ML) . Reasoning refers to the capacity of extracting knowledge; learning refers to the capacity of acquiring new knowledge from facts, automatically deriving a thesaurus of knowledge which makes the system capable of autonomous behavior in the real world. ML processes huge quantities of stored data to derive a model: the model enables us to make forecasts about values for future data. Forecasts are the more precise and the more accurate ones the more the model adheres to the real world. However, the real world may experience changes, thus drifting away from the original behavior, i.e., the one over which the model was initially defined. As a consequence, the performance of the model slumps. Two urgent needs arise: a) detect the performance slump of the model; b) decide when new training on more recent data is needed, to re-couple the model with the real world. The current paper aims at identifying a criterium to detect the need for re-coupling the model with the real world. As an application scenario, the paper considers data from blood examinations from [2]: data are then Published in the Workshop Proceedings of the EDBT/ICDT 2023 Joint Conference (March 28-March 31, 2023), Ioannina, Greece * Corresponding author. † These authors contributed equally. $ [oguzhan.avci@mail.polimi.it](oguzhan.avci@mail.polimi.it) (O. Avci); [giuseppe.pozzi@polimi.it](giuseppe.pozzi@polimi.it)[ ](giuseppe.pozzi@polimi.it)(G. Pozzi) 􀂀 https://www.deib.polimi.it/pozzi (G. Pozzi) 􀀚 —(O. Avci); 0000-0002-2828-862X (G. Pozzi) © 2023 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4 .0 International (CC BY 4 .0) . CEUR Workshop Proceedings ([CEUR-WS.org](CEUR-WS.org)) [http://ceur-ws.org](http://ceur-ws.org) [ISSN 1613-0073](ISSN 1613-0073) processed to build a predictive model, which classifies if the patient is affected by CoViD-19 or not. The paper is structured as follows: Section 2 describes the state of the art in ML; Section 3 focuses on the approach we propose here; Section 4 describes some details about the prototype implementation; Section 5 reports about the on-the-field deployment of the model; finally, Section 6 draws some concluding remarks. 2. Related Work This section describes some background issues and the state ofthe art on modelling and machine learning. 2.1. Background Given a huge quantity of data (the ground truth), ML enable us to build up a model which best fits t

FAQs

Why can a deployed machine learning model become obsolete?

Because the real world that generates the data can change over time, causing concept drift and degrading model performance.

How does the proposed approach decide when to update the model?

It uses a drift detector that monitors the deployed model’s performance in an online manner, detects a performance slump, and replaces the previous model with an up-to-date one.

What data and task are used to evaluate the method?

The experiments use 8642 hematochemical examinations from hospitalized patients, and the model predicts the RT-PCR test result for CoViD-19 (affected vs not affected).

What performance improvement does the method achieve?

It reaches an AUC of 0.794, about 6% better than offline and about 5% better than standard online binary classification techniques.

Aug 3, 2026

Upload by Hazel