| DOI | Resolve DOI: https://doi.org/10.1007/978-3-030-89906-6_53 |
|---|
| Author | Search for: Valdés, Julio J.1; Search for: Céspedes-González, Yaimara; Search for: Pou, Antonio |
|---|
| Affiliation | - National Research Council of Canada. Digital Technologies
|
|---|
| Format | Text, Book Chapter |
|---|
| Conference | FTC 2021 - Future Technologies Conference, Oct. 28-29, 2021, Vancouver, B.C., Canada [Held Virtually] |
|---|
| Subject | Process Mining; time series; inductive miner; machine learning; graph kernels |
|---|
| Abstract | Process Mining (PM) approaches for the analysis of time series bring interesting new perspectives because of the structural and behavioral information that could be extracted when the series are represented as event logs. This paper presents a Process Mining-based, supervised machine learning approach for constructing time series classifiers using graph kernels. A collection of time series of different kinds (normal, cyclic, with upward/downward trends and with upward/downward level shifts) is pre-processed by converting the real-valued series into event logs. Directly Follows Graphs (DFG) for activities and resources were discovered using the Inductive Miner algorithm and eight graph kernels were used for evaluating the similarities between the DFGs. These kernels were used in conjunction with class membership information for obtaining classification models in the form of support vector machines. High-performance classification models were found, particularly when using additively aggregated kernels coming from the combination of those obtained separately from activities and resources. The PM-based models have cross-validation mean accuracies comparable with those based on dynamic time warping, as well as top precision and recall metrics for most of the graph kernels. The results obtained are preliminary, but illustrate the potential of PM methods for time series analysis. |
|---|
| Publication date | 2021-10-24 |
|---|
| Publisher | Springer |
|---|
| In | |
|---|
| Series | |
|---|
| Language | English |
|---|
| Peer reviewed | Yes |
|---|
| Export citation | Export as RIS |
|---|
| Report a correction | Report a correction (opens in a new tab) |
|---|
| Record identifier | be21d2ee-1c24-4d16-844a-4b94cc8ca0cd |
|---|
| Record created | 2021-11-09 |
|---|
| Record modified | 2021-11-10 |
|---|
