-
Notifications
You must be signed in to change notification settings - Fork 4
Home
COMPLEX-IT is a web-based and downloadable RShiny app designed for non-specialists to interrogate their data with various tools from computational social science, such as Artificial Intelligence, micro-simulation, and predictive analytics. The app has a friendly and simple to use interface with quick introductions to the fundamental concepts, allowing users to quickly analyse their data; and accompanying video tutorials and instructions provide those who want it with richer details and information.
The applications of COMPLEX-IT are wide ranging, and with the appropriate dataset and an inquisitive mind could include:
- With limited resources as a local government authority, how effective would targeting education deprivation be in improving overall deprivation levels in our communities?
- As a social researcher with over 100 social experiment results, how can I group these into coherent groups?
- As a biologist, what are the relationships between the many variables in my large dataset relating to plant and animal responses to ecosystem stress?
In short, you don't need any technical expertise to start using COMPLEX-IT, all you need is a dataset you want to explore and a curious mind!
While the app has been designed with a chronological flow in mind, there is no need to follow all the steps if you don't want to! For example, you may already have a firm idea of the clusters within your dataset and feel little need for comparing this against the AI Self Organising Map. As such, there is nothing stopping you from going straight to investigating your variables with the systems mapping tab. Alternatively, you may only wish to use the scenario, policies, or interventions tab after having build your model with the k-means clustering and the SOM. Whatever analysis you do, your results will be generated and kept and can be downloaded in a report at the end of the app.
1) Build Database and Import Cases -- Upload your data, remove any columns, and preview data before performing any analysis. Each row can be considered a 'case' or instance, and each column a 'variable'. For example, in analysing local communities' deprivation scores, each row may be a town name and each column a facet of deprivation; or for a social reseracher each row may be a participant and the columns their scores on various recorded metrics.
2) Clustering Your Cases with K-Means -- How many clusters of similar groupings do you think there are in your data, based on any literature, prior experience, expertise, or just a hunch? Here you can test your data for any number of clusters and get feedback on how well each number of clusters best fits your data.
3) Use AI to Confirm Your Clusters -- Using a Self-Organising Map, AI will plot each case and try to place them into similar groups on a grid. Within the tab you can change settings to your liking such as how big the grid is or how many iterations the AI runs for, or you can use the default settings.
4) Compare and Visualise Your Results -- Here you can interrogate how well your clustering from step two matched up with what the AI found. For example, you can visualise your cases (and the clusters you put them in) compared to where the AI placed them on its SOM grid: do the two match up, and if not, why not?
5) Simulate Your Scenarios, Policies, or Interventions -- From the previous four steps you may be reasonably content in your model and what clusters your data can be sorted into. Here you can analyse how changing the variable values for your various clusters influences where they fall on the SOM grid. For example, as a local authority policy maker, you may be confident you can drop reported crime within a cluster of high deprivation communities by 20%, but will this be enough to move it away from the 'high-deprivation' area of the SOM grid? Moreover, how robust is this intervention to stress-testing -- what if you are confident you can get a 20% drop, but could forsee a drop of only 15%, or exceeding expectations and achieving a 25% drop. Within this tab you can explore these hypotheses and put them through such stress testing.
6) Use AI to Predict the Cluster Membership of New Cases --
7) Use Systems Mapping to Explore Your Cluster Variables -- Within your dataset there may be many variables (columns) per case (row). For example, within a dataset on communities' deprivation you may have variables for crime reported; income; unemployment; school outcomes; and many more. What are the relationships between all your variables, and what about the relationships between these variables within specific clusters too? Within this tab you can pull apart these questions.
8) Generate Your Report -- While you run your analyses in the various tabs, your most recent plots and statistical solutions are saved and can be downloaded in a report at the end of the app.