0
LightGBM vs. a Four-Week Average for Forecasting Bike-Share Rentals
https://towardsdatascience.com/lightgbm-vs-a-four-week-average-for-forecasting-bike-share-rentals/(towardsdatascience.com)An experiment was conducted to forecast hourly bike-share rentals for a 48-hour period using historical data from Washington, D.C.'s Capital Bikeshare. The study compares several models, including LightGBM, XGBoost, and Prophet, against a simple baseline of averaging the rental counts from the same hours over the prior four weeks. A significant part of the process involved careful data preparation, such as inferring zero rentals for missing hours and avoiding data leakage from future information. While the simple four-week average was highly competitive during development, the LightGBM model ultimately demonstrated a 19.6% lower error in a later evaluation period. The analysis highlights the importance of robust baselines and thorough model evaluation in time series forecasting.
0 points•by ogg•45 minutes ago
Comments (0)
No comments yet. Be the first to comment!
Have an account? Log in to join the discussion.