Thesis (BS Applied Mathematics) -- University of the Philippines Mindanao, 2010

This study developed a behavior pattern discovery (BPD) algorithm that takes as input a data set of binary matrices, called behavior matrices, and makes use of behavior prediction relationships. The proposed BPD algorithm is composed of two stages: (1) individual analysis of simple behaviors and (2) analysis of simple behaviors as a group. To analyze behaviors individually, the concept of shift was introduced. To analyze at least two behaviors as a group, a measure of association between behaviors, the predicting power, was proposed. The computation of predicting powers for different behavior combinations gives rise to behavior prediction relationships. These relationships indicate which behaviors significantly predict the state of which behaviors. It is important to note the proposed BPD algorithm mines both presence and absence patterns. The execution time of the original form of the BPD algorithm blows up as the dimensions of the data points increase. For this reason, two dimensionality-countering measures were incorporated to the original BPD algorithm: (1) behavior clustering and (2) determination of significant of expected behavior vectors of each behavior. The proposed BPD algorithm with the proposed dimensionality-countering methods was applied to the prey delivery instances (PDIs) 0f the Philippine eagle (P. jefferyi) for three different values of the shift value parameter, namely, 10, 20, and 30 minutes. The discovered behavior patterns were generally the same for the shift values 20 and 30 minutes, while those for the shift value 10 minutes were generally different from those of the other values. Two important realizations were made: (1) patterns of simple behaviors, when analyzed individually, are not necessarily the patterns of those behaviors when they are considered as part of group of behaviors and (2) relatively frequently occurring behaviors do not necessarily have discernable presence patterns.