The research is motivated by the need for hazard assessment in agriculture field. A small and highly-imbalanced dataset, in which negative instances heavily outnumber positive instances, is derived from a survey of secondary injuries induced by implementation of agriculture assistive technology which assists farmers with injuries or disabilities to continue farm-related work. Three data mining approaches are applied to the imbalanced dataset in order to discover patterns contributing to secondary injuries.

All of patterns discovered by the three approaches are compared according to three evaluation measurements: support, confidence and lift, and potentially most interesting patterns are found. Compared to graphical exploratory analysis which figures out causative factors by evaluating the single effects of attributes on the occurrence of secondary injuries, decision tree algorithm and subgroup discovery algorithms are able to find combinational factors by evaluating the interactive effects of attributes on the occurrence of secondary injuries. Graphical exploratory analysis is able to find patterns with highest support and subgroup discovery algorithms are good at finding high lift patterns.

In addition, the experimental analysis of applying subgroup discovery to our secondary injury dataset demonstrates subgroup discovery method's capability of dealing with imbalanced datasets. Therefore, identifying risk factors contributing to secondary injuries, as well as providing a useful alternative method (subgroup discovery) of dealing with small and highly-imbalanced datasets are important outcomes of this thesis.