DSC5104 Mini-Project Analyzing Resilience of Real-world Networks Background: In this project, we will look at bike-sharing data from Pronto Cycle Share, which was a public bike sharing company that operated in Seattle between 2014 and 2017. The company’s financial fortunes took a turn for the worse when it failed to secure funding in 2015, leading to its eventual shutdown in 2017. The dataset provided (see link below) for this mini-project spans information about the bike docking stations, the daily trips, as well as associated weather data from Oct 2014 to Aug 2016. Data source: Download the data from the following link: https://www.kaggle.com/pronto/cycle-share-dataset Please feel free to refer to the descriptive summary of the dataset on the Kaggle page, and also take a look at the various public kernels on the site for sample analyses and code snippets. If you are planning to reuse any code from these kernels, please remember to provide complete citation for the original user/source. You can make use of all the three data files in the dataset, for Task 1 and/or Task 2. Tasks: There are two tasks to be completed as part of this mini-group project. The first task is a defined task (similar to your Assignment 1), while the second task is open-ended and gives you the flexibility to uncover your own insights from the data (similar to your Term Project). Task 1: Network Resilience (10 marks) For this task, you will first have to construct a network from the given dataset. There are various kinds of networks that you can create using this data, and you are free to pick the one that you think has the most value/importance. (One example of a network could be one where nodes are bikes, and two bikes following a similar route in a particular month have a tie between them. The number of trips in that month could be the edge weight.) Also remember that the data is time-stamped, which means that you can construct a longitudinal network e.g. a set of networks, one for each week/month/quarter/year of the study period. Once the network is constructed, you are expected to analyze the “structural resilience” of this network. For the context of this study, let us define structural resilience as the tolerance of the network to loss of nodes, i.e. what happens to the connectivity of the network when nodes are gradually removed from this network. See the following paper for a good study and example of tolerance to errors and attacks in real- world networks: Albert, R., Jeong, H., & Barabási, A. L. (2000). Error and attack tolerance of complex networks. Nature, 406(6794), 378. https://arxiv.org/pdf/cond-mat/0008064.pdf There are broadly two ways by which nodes can be removed from the network: Type 1: Random deletion of nodes i.e. nodes are picked at random to be removed Type 2: Targeted deletion of nodes i.e. nodes are sorted based on some centrality measure, and higher centrality nodes are removed first. One of the major outcomes of node removal is that the network becomes less connected, and average path lengths start increasing. As a result, the network diameter also shows significant increase (see Fig.2 on Pg.12 in the paper cited above for an example). Implement the two types of node deletion strategies on your network, and plot the associated changes in diameter as a function of the fraction of nodes removed. Does the resulting plot resemble the Fig. 2 on Pg. 12 of the Albert et al. (2000) paper? Why do you (/don’t you) see this pattern? Can you think of any other important graph-level metric, apart from diameter, that can potentially measure resilience of the network to such errors and attacks? Compute this metric, and test if this metric also shows significant variation in response to these two types of node deletions. Repeat the above set of analyses for the network over time (e.g. for each month or quarter or year). Do you see any change in the resilience of the network over time? What patterns (if any) do you see from this temporal analysis? Task 2: Additional Analyses (10 marks) For this task, feel free to perform any additional analyses on the network that you have constructed as part of the previous task. The goal here is to uncover potentially interesting insights about the bike-sharing network that might offer some hints about why the bike sharing company did not perform as well as it had hoped to, and eventually shut down in 2017. Are you able to connect/generalize some of your findings from this task to relate to what other bike- sharing companies in Singapore and around the world are facing? Can you make useful recommendations to bike-sharing companies based on the network insights that you generated as part of Task 1 and Task 2 Submission guidelines Only a single submission per group is needed for this. Please restrict your project report to 4 pages (2 sheets), and submit this together with your code files as a single zip file latest by March 9th. Please mention your group number in the submission file name.
