Automated Machine Learning Pipeline for Modelling the Spectral Response of a Tapered Optical Fibre Temperature Sensor
Invited Talk
J M Cruz-Duarte1, J Korterik2, H L Offerhaus2, L de la Cruz-May3, L F Granados-Zambrano2, L Poveda-Wong4, D Jauregui-Vazquez5, E de la Cruz5, J A Alvarez-Chavez6
1 Centre Inria, Université of Lille, Lille, France
2 Optical Sciences group, University of Twente, Enschede, The Netherlands
3 Facultad de Ingenieria, Universidad Autonoma del Carmen, Campeche, Mexico
4 Escuela de Ingeniería Eléctrica, Universidad de Costa Rica, San José, Costa Rica
5 Center for Scientific Research and Higher Education at Ensenada, Esenada, Mexico
6 The Hague University of Applied Sciences, Delft, The Netherlands
Seminar: S8 — Fiber Optics
Abstract
An automated machine learning pipeline is proposed to model the spectral response of a tapered optical fibre temperature sensor. The approach incorporates physics- informed features derived from the sensor behaviour and applies Bayesian hyperparameter optimisation to improve model performance. Multiple regression strategies are explored, including neural network, tree-based, and symbolic regressors. This combination enables the pipeline to capture complex spectral patterns while maintaining model interpretability. As a result, the framework provides accurate predictions together with closed-form analytical expressions that describe the sensor response, facilitating both physical understanding and practical implementation in optical fibre sensing systems.