forecasting

Simple yet Effective Evolutions of Recursive Method without the need for Deep LearningPhoto by Halfcut Pokemon on UnsplashWhen working on a time series forecasting problem, a standard benchmarked approach is the recursive one. It can be easily used on top of any machine learning model, it requires low assumptions, and it’s easily explainable.Recursive forecasting consists in creating lagged features of the target series and fitting a machine learning model on them. When forecasting further steps in the future, the…

An Example with ThymeBoostPhoto by Hao Zhang on UnsplashOnline learning involves updating parameters with new data without re-fitting the entire process. For models which involve boosting, such as ThymeBoost, the ability to pick up where we left off would save valuable computation time.Luckily for us, the boosting procedure unlocks an easy path to updating parameters given new data. This is because the whole system is built to learn from residuals. Once we have new data, we simply need to predict for that new data and…

Showcasing the power of ensemble learning with decision trees and ARIMA modelsMeter Image — By Doris MorganIn this article, I am going to showcase the predictive power of ensemble learning for time series forecasting. Ensemble learning leads to models with higher predictive accuracy, a lower likelihood of overfitting, and a diverse set of predictions.I will be using a dataset provided by ASHRAE (The American Society of Heating, Refrigerating and Air-Conditioning Engineers) which has hourly meterage data on electricity,…

Introducing and implementing the ARIMA model in the AirPassengers datasetPhoto by Lukas on PexelsTime-series forecasting consists of making predictions in order to drive future strategic decision-making in a wide range of applications. Based on some of the terms introduced in this previous article such as trend and seasonality, this article focuses on the implementation of autoregressive-, integrated-, moving average-based models for time-series forecasting. Specifically, the structure unfolds as follows:Introduction to…

Ready to learn a quick-and-easy way to get traffic predictions for any amount of time in the future? Seriously. This article will show you how you can: Predict traffic changes, and maybe even let your boss know when periods of stagnation or negative growth are to be expected.What to expect during times of increased or decreased traffic, so you could tell if your declines are in line with predictions, or if something might be going wrong and traffic is declining more than it…

NOAAUS weather and climate forecasting is about to get a major upgrade. The National Oceanic and Atmospheric Administration (NOAA) this week flipped on two new supercomputers that are 3X faster than the agency's former supercomputing system.By this fall, the new HPE Cray supercomputers should deliver an improvement to the US Global Forecast System (GFS), a weather forecast model that generates data related to variables like temperatures, winds, precipitation, soil moisture and atmospheric ozone concentration. The new…

Building blocks for state-of-the-art modelsPhoto by Mourizal Zativa on UnsplashWhen talking about Time Series, people often tend to do it in a very practical data-oriented way: if you try and look up some definition, you may find expressions such as “a series of data points indexed (or listed or graphed) in time order” (Wikipedia), “a set of data collected at successive points in time or over successive periods of time” (Encyclopædia Britannica), “a sequence of data points that occur in successive order over some period…

Analyzing the trend and seasonality of the time-series to decompose the time-series and implement forecasting modelsPhoto by Lukas on PexelsTime-series forecasting is the task of making predictions based on time-stamped historical data. It involves building models to make observations and drive future decision-making in applications such as weather, engineering, economics, finance, or business forecasting, among others.This article is intended as an introduction to time-series forecasting. The structure unfolds as…

The truth is — there is no trendPhoto by Anna Kumpan on UnsplashTypically, when you see articles or guides on using models such as Xgboost or LightGBM for a time series problem, the author addresses the in-ability of a tree model to pick up a trend and forecast outside the bounds of the data. This is seen as a problem to overcome. But, it’s really a feature — not a bug. A feature that should only be turned off if we believe that our forecast needs to go out-of-bounds — not because a trend exists in our data.Consider what…

It’s all fun and games until your time-series is not Normal anymorePhoto by Heather Zabriskie on UnsplashARMA (AutoRegressive — Moving Average) models are arguably the most popular approach to time-series forecasting. Unfortunately, plain ARMA is made for Gaussian distributed data only. On the one hand, you can often still use ARMA by transforming the raw data. On the other hand, this typically makes probabilistic forecasts quite tedious.One approach to apply ARMA to non-Normal data are Copula models. Roughly, the latter…