Prototype of an IoT-based Activity Detection Service Generator with Evaluation Datasets

Here you find the data, scripts and Siddhi services used for the evaluation section of the paper "Online Detection of Process Activity Executions from IoT Sensors using Generated Event Processing Services" submitted to Future Generation Computer Systems (ISSN: 0167-739X). It is separated first by scenario (healthcare/smart-factory) and then into generation of services using one dataset and evaluation of those services using a range of other datasets, i.e., into siddhi-app-generation and app-evaluation. The datasets, accompanying scripts and other relevant files are in these subfolders.

Both scenarios contain the following files:

siddhi-app-generation/app_creation.py: Python script to generate the siddhi services

siddhi-app-generation/in/*: Input files for the generation (IoT data, expert annotations)

siddhi-app-generation/out/*: Output files from the generation (signatures, changes, siddhi services)

app-evaluation/iot/*.jsonl: IoT data used for the evaluation

app-evaluation/gt/*.xes: Ground truth data used for the evaluation

app-evaluation/det/*.xes: Detected activities used for the evaluation

app-evaluation/results: Some more detailed results of the evaluation

app-evaluation/*.py and app-evaluation/*.sh: Scripts to run the evaluation

An overview of the datasets can be seen below.

Data Overview

Smart Factory

Executed on a Fischertechnik Industry 9.0V smart factory model available at the University of St.Gallen

Input/Annotation/Generation

1 IoT log: 27.63 minutes, 0.5 Hz

13 activity types annotated

6 stations; no cross-station activities

14 sensors discretized

6 sensors ignored

Simulation/Evaluation

3 IoT logs: 73.99 min + 63.1 min + 41.35 min; roughly 3 hours; 0.5 Hz

395 activity instances

36 process instances

Smart Healthcare

Executed by 17 students of the Medical Master Programme at University of St. Gallen.

Input/Annotation/Generation

1 IoT log: 5.82 mins, 2 Hz

18 activity types annotated

2 stations; 3 cross-station activities

7 sensors discretized

3 sensors ignored general, some specific sensor ignores

Simulation/Evaluation

16 IoT logs: 65.31 minutes, 2 Hz

188 activity instances

16 process instances