Forecasting at Scale
By Sean J. Taylor, Benjamin Letham. 2018
2,851 citations recorded by the source corpus.
ABSTRACT Forecasting is a common data science task that helps organizations with capacity planning, goal setting, and anomaly detection. Despite its importance, there are serious challenges associated with producing reliable and high-quality forecasts—especially when there are a variety of time series and analysts with expertise in time series modeling are relatively rare. To address these challenges, we describe a practical approach to forecasting “at scale” that combines configurable models with analyst-in-the-loop performance analysis. We propose a modular regression model with interpretable parameters that can be intuitively adjusted by analysts with domain knowledge about the time series. We describe performance analyses to compare and evaluate forecasting procedures, and automatically flag forecasts for manual review and adjustment. Tools that help analysts to use their expertise most effectively enable reliable, practical forecasting of business time series.
Subfields: Management Science and Operations Research, Signal Processing, Time-Series & Forecasting
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Management Science and Operations Research
- Bernie Wang
- Tim Januschowski
- Jan Gasthaus
- Syama Sundar Rangapuram
- Rajat Sen
- Danielle C. Maddix
- Alex Smola
- Abhimanyu Das
In Signal Processing
- Bernie Wang
- Tim Januschowski
- Jan Gasthaus
- Syama Sundar Rangapuram
- Qiang Xu
- Ailing Zeng
- Danielle C. Maddix
- Rajat Sen
In Time-Series & Forecasting
- Bernie Wang
- Tim Januschowski
- Jan Gasthaus
- Danielle C. Maddix
- Rajat Sen
- Syama Sundar Rangapuram
- Alex Smola
- Bryan Lim
Citations in the corpus
- Deep Factors for Forecasting (2019, methodology): “…RNN-based probabilistic forecasting algorithm on the publicly available AWS SageMaker (Januschowski et al., 2018), MQ-RNN ( MR ), a sequence model that generates conditional predictive quantiles (Wen et al., 2017), and Prophet ( P ), a Bayesian structural time series model (Taylor & Letham, 2017).”
- A decoder-only foundation model for time-series forecasting (2023, positive): “Forecasting models can be categorized broadly into: (i) Local univariate models that include traditional methods like ARIMA, exponential smoothing (McKenzie, 1984) and non-autoregressive models like Prophet (Taylor & Letham, 2018).”
- Predictability of monthly temperature and precipitation using automatic time series forecasting methods (2018, methodology): “2017), while the Prophet method is implemented through the ‘prophet’ R package (Taylor and Letham 2017b). Subsequently, we present these seven methods, omitting the mathematical parts, since they are well documented in the literature, while highlighting important aspects from a practical point of vi”
- SCINet: Time Series Modeling and Forecasting with Sample Convolution and Interaction (2021, methodology): “…of SCINet in obtaining the trend and seasonality of time series for TSF. Univariate Time-series Forecasting on ETT : in this experimental setting, we bring several strong baseline methods for univariate forecasting into comparison, including ARIMA, Prophet [36], DeepAR [32] and N-Beats [29].”
- Transformers in Time-Series Analysis: A Tutorial (2022, methodology): “The model performed better than state-of-the-art (SOTA), including Autoregressive Integrated Moving Average (ARIMA) [85], Prophet [86], LSTMa [87], LSTnet [88], and DeepAR [89].”
- Deep Learning for Time Series Anomaly Detection: A Survey (2022, positive): “It is a forecasting model that integrates a GRU network using a time series decomposition method called Prophet [166], to enable robust learning on seasonal time series data without labels.”
- Deep Factors with Gaussian Processes for Forecasting (2018, methodology): “Table 1 compares with DeepAR (DA), a state-of-art RNN-based forecasting algorithm on the publicly available AWS SageMaker [7, 26] and Prophet (P), a Bayesian structural time series model [27].”
- Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting (2020, methodology): “We have selected 5 time-series forecasting methods as comparison, including ARIMA (Ariyo, Adewumi, and Ayo 2014), Prophet (Taylor and Letham 2018), LSTMa (Bahdanau, Cho, and Bengio 2015) and LSTnet (Lai et al. 2018) and DeepAR (Flunkert, Salinas, and Gasthaus 2017).”
- Pyraformer: Low-Complexity Pyramidal Attention for Long-Range Time Series Modeling and Forecasting (2022, background): “The first group involves ARIMA (Box & Jenkins, 1968) and Prophet (Taylor & Letham, 2018).”
- Darts: User-Friendly Modern Machine Learning for Time Series (2021, methodology): “, 2017), Theta (Assimakopoulos and Nikolopoulos, 2000), Prophet (Taylor and Letham, 2018), FFT-based forecasting, RNN models similar to DeepAR (Flunkert et al.”
- Fake news on Twitter during the 2016 U.S. presidential election (2019, methodology): “list (bottom) using the Bayesian time-series model proposed in (55).”
- Nonlinear Photovoltaic Degradation Rates: Modeling and Comparison Against Conventional Methods (2020, background): “PR trends and subsequent change points for estimating nonlinear RD in this work [21].”
- Bayesian Workflow (2020, background): “From another direction are menu-based packages such as Prophet (Taylor and Lethem, 2018) that allow users to put together models (in this case, for time series forecasting) from some set of building blocks.”
- ETSformer: Exponential Smoothing Transformers for Time-series Forecasting (2022, methodology): “Univariate baselines further include N-BEATS [24], DeepAR [26], ARIMA, Prophet [30], and AutoETS [3].”
- Are Transformers Effective for Time Series Forecasting? (2022, methodology): “…prediction accuracy improvements over traditional methods, in their experiments, all the compared (non-Transformer) baselines perform autoregressive or iterated multi-step (IMS) forecasting [1,2,22,24], which are known to suffer from significant error accumulation effects for the LTSF problem.”