InterSCity Experiment - Emulation Based On Anomaly Detection In Car Traffic

This repository contains our analysis and everything needed to run the experiment Emulation Based on Anomaly Detection In Car Traffic. In this experiment we evaluate actuation in a city traffic simulation via traffic boards, that alerts drivers about closed streets and are triggered via anomaly detection. In our context, we refer to events as close street events, that are configurable and predefined by us.

So, in this experiment, we:

• Simulate a city traffic

• During the simulation, a few events will affect the roads, closing them

• Since the roads behaviour are going to change, we will detect them as anomalies and trigger traffic boards

• Drivers that arrive at nodes containing triggered traffic boards will have a chance to recalculate their route, letting them plan a better route

We separated the experiment in two scenarios: in the first, that we called validation scenario, we used a fairly simple graph to validate our models and algorithms. In the second scenario, that we called sao paulo scenario, we used a reduced São Paulo map, with events in two clusters, and with traffic boards in high traffic edges. For each scenario we used a three phases separation, and in each phase we runned 20 simulations. The first phase is a simulation without events and without traffic boards, and was used as our baseline to find the thresholds to classify roads as anomalies. The second phase is a simulation with events and without traffic boards, and was expected to have the slowest travels in comparison to the others two phases. Finally, the third phase is a simulation with events and with traffic boards, were we used a small anomaly detection system (written in Apache Spark) to trigger the traffic boards.

Next, we describe the structure of this repository, which will help you to understand and reproduce the results.

img folder

This folder holds the images used in our Jupyter Notebooks. You can skip it! :)

running folder

This folder holds everything needed to reproduce the simulation and the scenarios. There you will find a README.md file that better describes the steps needed to rerun the scenarios.

scenario_1_validation folder

This folder holds the results of the first scenario simulated. In the scenario_1_validation/datasets folder you will find three folders, one for each phase of the scenario, and in each of these folders, an output.csv file, which aggregates the results of the 20 rounds simulated. Also, in each of these datasets folder there is a rounds folder, that stores the results of each round. In scenario_1_validation/inputs you will find the inputs used by the simulator for this scenario.

scenario_2_sao_paulo folder

This folder holds the results of the second scenario simulated. In the scenario_2_sao_paulo/datasets folder you will find three folders, one for each phase of the scenario, and in each of these folders, an output.csv file, which aggregates the results of the 20 rounds simulated. Also, in each of these datasetss folder there is a rounds folder, that stores the results of each round. In scenario_2_sao_paulo/inputs you will find the inputs used by the simulator for this scenario.

utils folder

This folder holds useful scripts and Jupyter Notebook that you might find useful if you want to reproduce everything that we used. For instance, here you can find a Jupyter Notebook that generates the map.xml file used by the simulator in the second scenario of the experiment.

Experiment Analysis

Our complete data analysis can be found in the Analysis.ipynb Jupyter Notebook file. You can open the file using Github itself to see our results and analysis or, to reproduce the results, you can open it with Jupyter Notebook and rerun the cells (we recommend reading this whole file before trying to run it). Also, if you have any questions, you can contact us in IRC (freenode#ccsl-usp).