Thesis title: Searching for optimal approaches in environmental classification and regression
The thesis presents a four-year investigation into the application of machine learning and model-based approaches for classifying images within the natural world and estimating pollutant concentrations. It is structured around three published papers and two manuscripts, each contributing to a comprehensive understanding of environmental analytics through statistical methodologies. The initial manuscript focuses on estimating atmospheric pollutants, specifically employing an ensemble modelling approach. This involves amalgamating individual predictive models, traditionally used in regression analysis, into a cohesive ensemble framework. Subsequent works explore the application of statistical models in the agricultural sector and the identification of forests and natural environments. Two papers delve into the potential of these algorithms to discriminate between categories within agricultural datasets. The remaining article and manuscript address the challenges of environmental and forest classification, proposing solutions for identifying natural habitats and assessing their health. From a statistical perspective, the thesis initially showcases the development of an ensemble model in a regression setting and then shifts the focus to evaluating the efficacy of machine learning algorithms against traditional model-based classification methods for identifying frequent and rare categories in these fields.