Ngakan Krisnawijaya

By introducing smart farming systems, farm management decisions can be enhanced based on solid data and good knowledge. Smart farming systems incorporate various emerging technologies to transform raw data into valuable information. When the required information is provided to the farmers through digital applications, they can learn and combine the information with their farm experiences to make better decisions. Smart farming systems are beneficial in order to improve daily farm management. Due to the enhancement of data management and analytics platforms, smart farming systems have become more powerful in providing information based on comprehensive analysis.

Although smart farming systems provide several benefits, their implementation in practice currently lags behind expectations due to a lack of proper infrastructure. Therefore, a system architecture that is adaptive, maintainable, and robust is needed to cope with different implementations of smart farming systems. Thus, the objective of this research is to analyze and develop a generic reference architecture model for the data management and analytics platform for smart farming systems.

In Chapter 2, we explore how data management and analytics in smart farming can assist countries still using conventional farming methods to adopt more precise, data-driven approaches. By doing so, we believe that these countries can improve their food production. We conducted a systematic literature review to gather insights into the technologies, data properties, architecture, and algorithms used in agricultural data analytics platforms (detailed in Chapter 2). We identify 15 stakeholders, 33 features, and 34 obstacles of the adopted data analytics platforms.

Based on the SLR results, we propose a reference architecture for data management in smart farming provided in Chapter 3. The architecture was developed through domain analysis and architecture modeling approaches. From the domain analysis, the common and variant features and modules of the smart farming system were studied. As a result, a blueprint representing family features across various smart farming domains was derived.

Furthermore, in Chapter 4, we involve the stakeholders in selecting a cloud-computing-based data analytics platform for smart farming. Selecting a feasible cloud platform is challenging since the decision-makers need to consider many factors and criteria. In addition, a challenge when working in a diverse group with stakeholders from various backgrounds is dealing with knowledge gaps, different needs, expectations and opinions. Thus, Chapter 4 provides a systematic way to deal with the aforementioned problems. We use integrated multi-criteria decision analysis (MCDA) approaches, using the Analytic Hierarchy Process (AHP) and Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) as an evaluation framework and model evaluation. The result of the study demonstrates that our proposed model can manage different criteria and help decision-makers choose the cloud platform based on the criteria.

FAIR data principles are becoming proper rules in the realm of data management systems. These principles provide guidance for developing data management systems. They help to manage the data for data sharing and reusing among stakeholders across different domains, backgrounds and disciplines. Chapter 5 shows the investigation of the implementation of FAIR principles in data management systems. The study aims to provide knowledge regarding FAIR data principles within the data management system of an interdisciplinary agricultural project. The findings provide valuable guidance on best practices and pitfalls to avoid when implementing data management systems.

One of the data analytics purposes in agricultural systems that we obtained from the literature is to detect disease. As a comprehensive analysis of how a disease can be detected from the raw data and how data management and analytics processes in smart farming systems can be applied, analytical work is provided in Chapter 6 based on a real-world data set from a dairy cooperative. A machine learning model, XGBoost, was used along with our proposed noise detection algorithm and several imbalanced data management methods to detect FMD cases. Our model shows the importance of data preparation before analysis phase, which resulted in the model's good performance in identifying FMD cases.

Based on the aforementioned works, we have learned and aimed to share that developing a smart farming system is not a straightforward task. A proper architecture is needed to show gross-level design and the quality of the system. Furthermore, we provide a systematic way to find a consensus when working in diverse groups and to implement FAIR principles that are important when collaborating with others from different domains or organizations. We have also demonstrated that proper data management can help enhance the analytics model to identify a disease. From our research, we noticed that the role of the stakeholders was significant in achieving the goals of the smart farming systems, from the design process to the implementation phase. In addition, the quality of data fed into the smart farming system must be really carefully considered and processed, as it plays a significant role in determining the reliability, robustness, and trustworthiness of the system's results.