Explainable boosted linear regression for time series forecasting

Time series forecasting involves collecting and analyzing past observations to develop a model extrapolate such into the future. Forecasting of future events is important in many fields support decision making as it contributes reducing uncertainty. We propose explainable boosted linear regression (EBLR) algorithm for time forecasting, which an iterative method that starts with base model, explains model’s errors through trees. At each iteration, path leading highest error added new variable model. In this regard, our approach can be considered improvement over general models since enables incorporating nonlinear features by residual explanation. More importantly, use single rule most access interpretable results. The proposed extends probabilistic generating prediction intervals based on empirical distribution. conduct detailed numerical study EBLR compare against various other approaches. observe substantially improves performance extracted features, provide comparable well established interpretability predictions high predictive accuracy makes promising forecasting.