t-SNE: A study on reducing the dimensionality of hyperspectral data for the regression problem of estimating oenological parameters

• DOI: 10.1016/j.aiia.2023.02.003
• Publication Date: 2023-03-06T21:46:04Z
• Authors: Rui Silva; Pedro Melo-Pinto
• Keywords: Kernel (algebra)

In recent years there is a growing importance in using machine learning techniques to improve procedures precision agriculture: this work we perform study on models capable of predicting oenological parameters from hyperspectral images wine grape berries, specially relevant topic boost production tasks for winemakers. Specifically, explore the capabilities novel technique mostly used visualization, t-Distributed Stochastic Neighbor Embedding (t-SNE), reducing dimensionality highly complex data and compare its performance with Principal Component Analysis (PCA) method, which despite introduction many nonlinear reduction over years, had achieved best results real-world across several studies literature. Additionally potential Kernel t-SNE, an extension t-SNE method that allows usage streaming or online scenarios. Our show that, direct comparison, achieves better metrics than PCA most sets regressor (Support Vector Regression, SVR) performs reduced features as inputs, accomplishing predictions lower error rates. Comparing current literature, our shallow model paired either par those reported, even competing more advanced use deep techniques, should propel require reduction.