1. Recap 2 Tune again Modelling Retuning 2.1 Retune Random Forest 2.2 Retune Prophet boost 2.3 Performance (after retuning) 3 Ensemble 3.1 Peformance (ensemble) 4 Performance (individual levels) Hospital Cluster level National level 5 The future Hospital Cluster National 6 KIV Plans Errors 1. Recap The aim of this series of blog is to predict monthly admissions to Singapore public acute adult hospitals.
1 Intro 2 Cross validation Metrics 3. Pre-processing 3.1 Base recipe 3.2 Spline recipe 3.3 Prophet boost recipe 4 Modelling 4.1 GLM 4.2 MARS 4.3 RF 4.4 XGB 4.5 Prophet boost 5. Evaluation Conclusion 1 Intro The aim of this series of blog is to predict monthly admissions to Singapore public acute adult hospitals. EDA for the dataset was explored in past posts ( part 1 ; part 2 ).
1 Intro 2 Data wrangling 2.1 Long format with aggregated values 2.2 Extend into the future 2.3 External regressor 2.3.1 Lags and rolling lags 2.3.2 Covid 2.3.3 Time series features 3 Splitting 4 Pre-processing recipes Pre-processing order 5. Modelling Workflow 6. Evaluate 6.1 Evaluate against the training set What’s inside the calibrated table 6.1 Evaluate with cross validation 8 Conclusion 1 Intro The aim of this series of blog is to predict monthly admissions to Singapore public acute adult hospitals.
Introduction 1. Lags 1.1 Auto-correlation (ACF) 1.2 Partial Autocorrelation (PACF) 2. Correlation of features Hospital level Cluster level National level All levels Correlation findings 2.1 PCA of features 2.1 Variation captured 2.2 Time series features contributing to PC1 2.3 Distribution of hosptials in PC plane Conclusion Errors Introduction The aim of this series of blog is to predict monthly admissions to Singapore public acute adult hospitals.
Introduction Visualization 1. Trend 2. Seasonality Trend and seasonality 3. Anomaly Conclusion Introduction The aim of this series of blogs is to do time series forecasting with libraries that conform to tidyverse principles and there are two of these time series meta-packages
modeltime which is created to be the time series equivalent of tidymodels fpp3 which is created to do tidy time series and has been nicknamed the tidyverts.