How do the temporal model parameters (backward window, prediction window, and gap) influence the model's performance?
Context: #6
Deliverables
Backward window: we want to know if increasing the size of the backward window improves overall performance, or on the contrary adds too much noise.
Prediction window: we want to measure the performance of prediction algorithms with different prediction windows, ranging from a couple months to several years (e.g. 10 years).
Gap: changing the size of the gap will help us determine
For each of these, statistics should be collected to understand the temporal distribution of data for patients (e.g. how many of them actually have a history of 10+ years)
Make these tests for at least 2 diseases and 2 data sources. Consider introducing a metric for the evolution of the predictive power of features depending on the temporal windows (look into temporal cross-validation, information gain over time, time-dependent ROC curves, etc.)
Approaches
Experiments needed to address the question
[ ]
Understanding and Exploration
A (single or few) disease(s) will need to be selected for this case study.
Problem Statement
How do the temporal model parameters (backward window, prediction window, and gap) influence the model's performance?
Context: #6
Deliverables
Approaches
Experiments needed to address the question
Understanding and Exploration