Files

04_R2_Manuscript with author details.pdf
  • Open Access
  • Adobe PDF
  • 3.58 MB
Download

Details

Authors
Abstract
Supervised machine learning (ML) is emerging as a powerful tool for handling complex multivariate data sets and capturing nonlinear relationships between predictors and outcomes. Unlike traditional statistical techniques that only enable the identification of predefined functional relationships, supervised ML methods can robustly detect intricate nonlinear interactions without explicit assumptions. The capacity for integrating diverse predictors makes these tools particularly suitable for challenges for which traditional modeling shows limitations. Prediction of discomfort due to glare is one such case, where current models often fail to explain variance in data. These models mostly focus on objective factors related to the scene, while neglecting subjective factors that can vary between- and within-users. These limitations contribute to significant residual variance between model-predicted and reported discomfort scores, which cannot be explained without accounting for inter- and intra-individual differences. To improve these predictions, this paper proposes a method based on supervised machine learning techniques where multiple predictors and their nonlinear relationships are integrated towards improving model accuracy and robustness. Via preprocessing, features of a data set collected from a lighting experiment were refined, and prediction models were developed using eight state-of-the-art ML algorithms. Through inverse regression, the best performing model was further deployed for optimizing the characteristics of delivered lighting, ensuring that user perception was maintained within a target zone when all other variables were constrained. Our method demonstrates a practical approach for data handling, multivariate nonlinear modeling, and user-centered optimization of delivered lighting, with potential application for problems beyond prediction of discomfort due to glare.
Affiliations

Citations

  • APA
  • Chicago
  • FWB

Maskarenj, M., & Altomonte, S. (2026). Supervised Machine Learning Methods for the Prediction of Discomfort due to Glare. Leukos (Print), 1(27), 1-27. https://doi.org/10.1080/15502724.2025.2609655 (Original work published 2026)