In our lives it is a more and more common situation where our luggage shall be examined. Earlier, the luggage control happened mainly at border crossing points or at the entrance of unique and highly visited tourist facilities (e.g. Saint Peter's Basilica), or maybe at penal institutions. However, today the safety standards more often restrict the allowed objects to a place or to an institute. On the other hand, more sophisticated technological systems are required to detect the presence of prohibited substances. One possible way is the odour-based recognition of objects. This problem can be solved using data mining, which is a very popular and interdisciplinary field of computer science.
In this thesis I build data mining models using available large training data sets. The model recognizes the patterns in the evaluation data sets, answering the question of when and how the objects were shown to the odour detection device. In this particular case, pattern recognition means classification and the used data mining algorithm is the SVM. In the test phase this algorithm achieves 5–10% better results over the others.
During the study, I go through the steps of the CRISP-DM data mining methodology, supplemented by the description of the theory of classification algorithms. In addition, I confirm the correctness of the chosen algorithm and the effectiveness of the final model-set by presenting logical test cases. The thesis ends with the evaluation of the results, which shows that the correct pattern recognition ability of the final model-set is over 90%.