We addressed the issue of rhythmic data analysis. We presented various approaches for analysis of rhythmic data. We specifically focused on the Cosinor model, its variations, and its connection with linear regression. We introduced three models (Ridge, Lasso, ElasticNet) for determining the optimal number of components using the Cosinor model. We thoroughly tested these models on synthetic data according to different parameters and demonstrated their applicability on real data. We found that the Lasso and ElasticNet models are most suitable for determining the optimal number of periods, as they most aggressively remove period components. Parameterization is also important in all models, as different parameter values significantly impact the results.
|
