Creative Innovation: Models and Analysis

Due to the number and diverse amount of data being analyzed for this effort, our method requires several different approaches. These analyses can be found in our Data Analysis folder.

Create an adjacency matrix containing the nations of the world.

An unsupervised model, in which nations with similar levels of innovation will be grouped together.

A supervised learning approach with predictors for innovation.

The results of our supervised and unsupervised models are visualized below. Users can use the dropdowns below to compare the results within each model type, as well as compare results between the supervised and unsupervised models.

Unsupervised Model: K-means Clustering

K-means clustering was used for the unsupervised model. We hypothesized that innovation is influenced by both creative and non-creative (economic) factors. However, to determine which combination of factor types (creative only, non-creative only, or both) was the best set of inputs for innovation, we ran separate clustering models for each of these three conditions, allowing the user to compare between each set of factors.

The dropdown can be used to toggle between the three conditions, while the slider allows for variation in the number of clusters.

Select Analysis Output

Supervised Model: Regressions

Regressions that incorporated feature selection were used for the supervised models, which was our attempt to assign an innovation score for each country. Predictor variables consisted of both creative and non-creative (economic) factors, while the global innovation index was used as a proxy for an innovation response variable ("original innovation index").

Since the integrated dataset included dozens of columns, feature selection was a key component in our models. We explored a variety of models but ultimately decided to visualize the results from the:
  • recursive feature elimination (RFE) linear regression model ("rfe prediction")
  • lasso linear regression model ("lasso prediction")
The dropdown can be used to toggle between the different innovation scores that were both modeled by the regression models and provided by the global innovation index.

Select Analysis Output