tspredit
Time Series Prediction with Integrated Tuning
Time series prediction is a critical task in data analysis, requiring not only the selection of appropriate models, but also suitable data preprocessing and tuning strategies.
TSPredIT (Time Series Prediction with Integrated Tuning) is a framework that provides a seamless integration of data preprocessing, decomposition, model training, hyperparameter optimization, and evaluation.
Unlike other frameworks, TSPredIT emphasizes the co-optimization of both preprocessing and modeling steps, improving predictive performance.
It supports a variety of statistical and machine learning models, filtering techniques, outlier detection, data augmentation, and ensemble strategies.
More information is available in Salles et al.
TSPredIT
TSPredIT (Time Series Prediction with Integrated Tuning) is an R framework for time series forecasting workflows. Built on top of DAL Toolbox, it helps users move from a raw series to complete and reproducible forecasting pipelines.
The package treats forecasting as a sequence of explicit decisions: how to represent a series, how to split it while preserving temporal order, whether to filter noise, whether to augment training windows, how to normalize values, which predictor to use, and how to compare forecasting protocols. This makes the package useful for teaching, experimentation, benchmark studies, and applied time series projects.
Current package version in this repository: 2.0.707.
Installation
The stable version is available on CRAN:
https://CRAN.R-project.org/package=tspredit
install.packages("tspredit")
The development version is available on GitHub:
https://github.com/cefet-rj-dal/tspredit
library(devtools)
devtools::install_github("cefet-rj-dal/tspredit", force = TRUE, upgrade = "never")
Documentation
Documentation and examples are available in the package site and in the repository:
The documentation is organized around two complementary entry points:
- a guided tutorial track for readers who want to learn the forecasting workflow step by step
- thematic example collections for readers who want to inspect one stage of the forecasting pipeline at a time
If you are new to tspredit, start with the tutorials. If you already
know the package structure, the thematic collections provide focused
examples by pipeline stage.
Guided Tutorial Track
The tutorials are part of the examples collection. They form a 10-part sequence that starts with ARIMA forecasting protocols, then builds the sliding-window MLP pipeline piece by piece, and ends with model comparison and integrated tuning.
The sequence is cumulative. Each tutorial introduces one main decision in a time series forecasting study and keeps the code close to that learning objective.
Thematic Example Collections
The examples collection includes thematic subcollections:
- Time-series data utilities - build
ts_data, project windows into inputs and targets, and create train/test splits that preserve temporal order. - Datasets - inspect the packaged datasets documented in
R/data.RandR/tspredbench.R, one dataset at a time. - Filtering - compare identity baselines, smoothing methods, robust filters, decomposition methods, state-space filters, and seasonal adjustments.
- Lag mapping - study how different lag-selection rules decide which past observations are exposed to the predictor.
- Augmentation - study when synthetic windows help, from simple perturbations to recency-aware transformations.
- Normalization - inspect how scale, drift, adaptive transformations, and differencing affect the signal seen by the predictor.
- Prediction - move from classical baselines to machine-learning and neural forecasting models, then to tuning.
- Multivariate forecasting - extend the pipeline to target-centered
multivariate workflows by combining one model for
ywith auxiliary-variable pipelines. - Custom extensions - learn how to add custom predictors, filters, augmentations, and normalization methods without breaking the package contract.
Main Capabilities
- Time-aware data representation and train/test splitting.
- One-step, rolling-origin, and multi-step forecasting protocols.
- Classical, machine-learning, and neural forecasting models, including ARIMA, KNN, SVM, random forest, ELM, MLP, Conv1D, LSTM, DARIMA, and WARMA workflows.
- Filtering, smoothing, seasonal adjustment, and decomposition-based preprocessing.
- Lag mapping strategies based on recency, spacing, geometry, autocorrelation, partial autocorrelation, seasonal behavior, peaks, blocks, mutual information, and supervised feature selection.
- Data augmentation and normalization strategies for time series forecasting.
- Integrated tuning for input size, preprocessing choices, augmentation choices, and model hyperparameters.
- Support for multivariate forecasting workflows.
Course Material
The public course material includes a compact slide sequence that introduces TSPredIT and demonstrates the main package workflows: overview, tutorial, data utilities, datasets, filtering, augmentation, normalization, prediction, and custom extensions.
Related DAL Projects
Playlist
Bug Reports and Feature Requests
Please report bugs, questions, and feature requests at: