ShipCohortStudy is a project aimed at using Interpretable Machine Learning to understand participant evolution in longitudinal cohort study data. ShipStudyApi is a restful service that exposes http endpoints to visualize the results of rulefit model built on SHIP dataset. Additionally, evoxploit package is used to augment the dataset with evolutionary features (Niemann et al., 2015) and rules extracted from building random forest on the dataset (Friedman et al., 2008). The feature selection is then performed using Least Absolute Shrinkage and Selection Operator(LASSO) to obtain minimal set of important features (Friedman et al., 2009). The api caters to a visualization dashboard that explores the model results dynamically and interactively.
Checkout Dashboard Here
- Measure the influence of each evolution feature on the predicted class.
- Evaluate the merit of an evolution feature towards classification accuracy.
- Analyse the minimal change in the participant such that the predicted class label changes.
- Identify the minimal set of evolution features that would result in better predictive performance.
- Navigate to cloned repository / project directory and run the following command
Rscript src/R/app.R <input_dataset_path>- The model building might take from 15 to 30 mins depending on the machine performance.
- After running the above command, the api service will be available from http://localhost:3000/__swagger__/
- (Niemann et al., 2015) Uli Niemann, Tommy Hielscher, Myra Spiliopoulou, Henry Völzke, and Jens-Peter Kühn. "Can we classify the participants of a longitudinal epidemiological study from their previous evolution?" Proc. of IEEE Computer-Based Medical Systems, 121-126, 2015.
- (Friedman et al., 2008) Friedman, Jerome H.; Popescu, Bogdan E. Predictive learning via rule ensembles. Ann. Appl. Stat. 2 (2008), no. 3, 916--954. doi:10.1214/07-AOAS148.
- (Friedman et al., 2009) Friedman, Jerome, Trevor Hastie, and Robert Tibshirani. “The elements of statistical learning”.