Finding Patterns in Vehicle Diagnostic Trouble Codes : A data mining study applying associative classification

Sammanfattning: In Scania vehicles, Diagnostic Trouble Codes (DTCs) are collected while driving, later on loaded into a central database when visiting a workshop. These DTCs are statistically used to analyse vehicles’ health statuses, which is why correctness in data is desirable. In workshops DTCs can however occur due to work and tests. Nevertheless are they loaded into the database without any notification. In order to perform an accurate analysis of the vehicle health status it would be desirable if such DTCs could be found and removed. The thesis has examined if this is possible by searching for patterns in DTCs, indicating whether the DTCs are generated in a workshop or not. Due to its easy interpretable outcome an Associative Classification method was used with the aim of categorising data. The classifier was built applying well-known algorithms and then two classification algorithms were developed to fit the data structure when labelling new data. The final classifier performed with an accuracy above 80 percent where no distinctive differences between the two algorithms could be found. Hardly 50 percent of all workshop DTCs were however found. The conclusion is that either do patterns in workshop DTCs only occur in 50 percent of the cases, or the classifier can only detect 50 percent of them. The patterns found could confirm previous knowledge regarding workshop generated DTCs as well as provide Scania with new information.