Predictive Analytics Models and Tools for Decision Support of Stakeholders in Digital Agriculture

Recent advances in information and communication technologies (ICT) have led the agricultural sector to a new technological paradigm, known as Digital Agriculture. The carrier of this digitization is driven by the massive wealth of “bio” and “info” data produced by ICT. Therefore, Big Data represent a foothold for the development of technologies related to decision support that stakeholders must take to optimize their productions. Strategic and necessary tools, to face the challenges dictated by climate changes combined with the rising global demand for food, which recall to improve productivity, increase food and nutrition security, and promote sustainable agriculture. Sustainability achievable through the development of models and tools equipped with Artificial Intelligence. Intelligent support along the entire agri-food chain presents a series of advantages, optimizing production and quality efficiency, reducing business costs, optimizing inputs to minimize environmental impacts, and generating business opportunities in a revitalized sector. The thesis deals with the study, implementation and evaluation of predictive models that perform a dynamic and cross analysis of multidimensional and multisensor data from different distributed sources. The model predictions aim to support stakeholders in the prevention and diagnosis of crop diseases. Analyses related to climatic, environmental and cultural aspects allow to monitor and predict the crop risk situations in the short term, and to take timely actions weighted to the needs of the monitored plot. The scientific contribution is articulated on different levels: from the construction of a dataset, to the implementation and validation of predictive models in local conditions, to the development of a prototype of Decision Support System, called LANDS (Laore Architecture Network Development for Sardinia), to the developlment of a toolbox to diagnose leaf diseases, called LeafBox. Through our studies, we present opportunities and challenges introduced by Machine Learning for the above goals, a relevant and timely topic in literature. Overall, we believe that our studies on machine learning-based solutions explore a promising direction, and can be used as a reference guidelines for future research and applications for both academia and industry sectors. The thesis outputs are publicly available at www.francescamalloci.com