Hierarchical Cluster Analysis: A reliable tool allowing more detailed (regional) traceability investigations

Abstract

We are presenting an artificial intelligence application designed to perform a reliable recognition of the geographical origin of horticultural products. The system allows more detailed traceability investigations than those required by the present general and specific standards imposed by the European Community legislation. The classification is performed by using an unsupervised pattern recognition technique, i.e. Hierarchical Cluster Analysis. The efficiency of the system is illustrated for dill (Anethum gruveoles), which is one of the most popular spice in Europe. Dill is also used for its digestive, antispasmodic, antiinflammatory, diuretic and antioxidant properties. The knowledge base includes physico-chemical information about dill samples originating from four neighboring regions of Romania and of the Republic of Moldova. The inference engine assigns the class identity (region of origin) based on agglomerative clustering. The results show that the system is a remarkably reliable tool for indepth traceability investigations. It clearly discriminates dill samples originating from closely located regions, which are characterized by quite similar pedo-climatic conditions. The human-machine interface is user-friendly, allowing the system to be easily used even by non-specialists. The sensitivity of selectivity of the system is discussed in comparison with those obtained by using Principal Component Analysis.