Tools for machine-learning based time-series prediction

This R package provides tools for machine-learning based time-series prediction. This includes preprocessing, creating matrices of lags, building a pipeline, and specialized cross-validation.

Installation

You can install the latest version from Github:

if (!require("devtools")) install.packages("devtools")
devtools::install_github("fnoorian/mltsp")

Usage

This package includes two vignettes:

• NARX model guide: Simple solution for non-linear auto-regressive forecasting.
• Pipeline model guide: Introduces pipeline model and cross-validation tools.

NARX model example

To build a non-linear auto-regressive model for lh (Luteinizing Hormone in Blood Samples, datasets package) with

• differencing order d = 0,
• auto-regressive order p = 3,
• seasonal frequency freq = 1,
• seasonal differencing order D =0,
• seasonal order P = 0,
• svm learner from package e1071,
• No external regressor

use:

library("mltsp")
library("e1071")

spec = build_narx(svm, p = 3, d = 0, P = 0, D = 0, freq = 1)
model = narx(spec, lh, xreg = NULL)

These steps can be combined into one (and default values can used):

model = narx(lh, svm, p = 3)

To predict 5 steps ahead, simply use:

forecast(model, h = 5)

Output is compatible with package forecast.

Pipeline model example

To build a pipeline for predicting lh (Luteinizing Hormone in Blood Samples from datasets package) with

• Preprocessing: differencing with order automatically determined
• Features: three lags of input data
• Learner: a linear model

use:

library("mltsp")

pp <- list(list("diff", "auto"))
fx <- function(x) cbind(x, lag_windows(x, p = 3))
ln <- SimpleLM
fcster <- mltsp_forecaster(pp, fx, ln)

and then apply function fcster to data:

fcster(lh, h = 5)

Cross validation example

To get cross-validation score for fcster function for a horizon of 5 steps, with training on at least 10 data points, use:

ts_crossval(lh, fcster, horizon = 5, initial_window = 10)

License and contact information

Copyright 2016 Farzad Noorian [email protected], Richard Davis [email protected], Anthony Mihirana de Silva [email protected], Lei Li [email protected].

Some functionality have been taken from GPL package caret. For a list of other functions dynamically imported from other R packages, see NAMESPACE file.

This package is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 2 of the License, or (at your option) any later version.

This package is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this package. If not, see http://www.gnu.org/licenses/.