Smart Predictive Agriculture Based on Data Science

Mancer, M’hamed (2025) Smart Predictive Agriculture Based on Data Science. Doctoral thesis, Université Mohamed Khider (Biskra - Algérie).

Text thesis-Mancer-final.pdf Download (10MB)

Abstract

Smart agriculture integrates digital technologies, sensors, the Internet of Things, big data, and artificial intelligence to transform traditional farming into precision-oriented and datadriven systems. These systems aim to improve productivity while making better use of resources. At the beginning of each growing season, farmers must make decisions that guide the success of the entire production cycle. The most important of these choices is deciding which crops to plant and how to divide land among them. This choice influences all later activities, such as planning the planting schedule, preparing the soil, and organizing the use of inputs. Because of its importance, crop selection is often described as the first step in farm planning. The first contribution of this thesis responds to this problem by introducing an interpretable crop selection system. The system integrates SHAP-based explanations to show how soil properties and climate conditions affect each recommendation. It combines strong predictive ability with clear explanations, offering a practical tool that farmers and advisors can use with greater trust. After the crop has been chosen, the next important question is “how much to expect.” Accurate yield forecasting allows farmers to organize inputs, schedule labor, manage uncertainty, and prepare for market activities. The second contribution of this thesis addresses this by designing a stacked ensemble learning framework, developed with greenhouse tomato production as a case study. It delivers accurate daily yield forecasts and achieves better results than standard regression methods, providing a reliable decision-support tool for greenhouse management. Since both crop selection and yield forecasting depend on the quality of agricultural data, the third contribution focuses on how this data can be kept secure, reliable, and trustworthy. To achieve this, a blockchain-based approach is proposed that integrates encryption, distributed file storage, and smart contracts. The approach ensures data traceability, confidentiality, and tamper-resistance.

Actions (login required)

View Item