Title:	Drift Adaptable Models
Version:	1.2.727
Description:	In streaming data analysis, it is crucial to detect significant shifts in the data distribution or the accuracy of predictive models over time, a phenomenon known as concept drift. The package aims to identify when concept drift occurs and provide methodologies for adapting models in non-stationary environments. It offers a range of state-of-the-art techniques for detecting concept drift and maintaining model performance. Additionally, the package provides tools for adapting models in response to these changes, ensuring continuous and accurate predictions in dynamic contexts. Methods for concept drift detection are described in Tavares (2022) <doi:10.1007/s12530-021-09415-z>.
License:	MIT + file LICENSE
URL:	https://cefet-rj-dal.github.io/heimdall/, https://github.com/cefet-rj-dal/heimdall
Encoding:	UTF-8
RoxygenNote:	7.3.3
Imports:	stats, caret, daltoolbox, ggplot2, Metrics, reticulate, pROC, car
Config/reticulate:	list( packages = list( list(package = "scipy"), list(package = "torch"), list(package = "pandas"), list(package = "numpy"), list(package = "matplotlib"), list(package = "scikit-learn") ) )
NeedsCompilation:	no
Packaged:	2026-03-10 18:09:07 UTC; gpca
Author:	Lucas Tavares [aut], Leonardo Carvalho [aut], Rodrigo Machado [aut], Diego Carvalho [ctb], Esther Pacitti [ctb], Fabio Porto [ctb], Eduardo Ogasawara [aut, ths, cre], CEFET/RJ [cph]
Maintainer:	Eduardo Ogasawara <eogasawara@ieee.org>
Repository:	CRAN
Date/Publication:	2026-03-10 18:40:02 UTC

ADWIN method

Description

ADWIN (Adaptive Windowing) is a sequential change detector that maintains a variable-length window and tests whether the means of two subwindows differ significantly. In this package, the implementation is primarily used for virtual concept drift when it monitors a numeric feature stream, although the same mechanism can also detect real concept drift if applied to an error or loss stream. The theoretical basis follows Bifet and Gavaldà (2007) doi:10.1137/1.9781611972771.42.

Usage

dfr_adwin(target_feat = NULL, delta = 2e-05)

Arguments

Value

dfr_adwin object

References

Bifet, A., and Gavaldà, R. (2007). Learning from time-changing data with adaptive windowing. In Proceedings of the 2007 SIAM International Conference on Data Mining, 443-448. doi:10.1137/1.9781611972771.42

Examples

#Use the same example of dfr_cumsum changing the constructor to:
#model <- dfr_adwin(target_feat='serie')

Autoencoder-Based Drift Detection method

Description

AEDD is an unsupervised multivariate detector that compares reconstruction errors produced by an autoencoder on reference and recent windows. Because it monitors changes in the input distribution rather than classifier performance, this implementation is primarily aimed at virtual concept drift. The method follows Kaminskyi, Li, and Muller (2022) doi:10.1109/ICDMW58026.2022.00109.

Usage

dfr_aedd(
  encoding_size,
  ae_class = autoenc_encode_decode,
  batch_size = 32,
  num_epochs = 1000,
  learning_rate = 0.001,
  window_size = 100,
  monitoring_step = 1700,
  criteria = "mann_whitney",
  alpha = 0.01,
  reporting = FALSE
)

Arguments

Value

dfr_aedd object

References

Kaminskyi, D., Li, B., and Muller, E. (2022). Reconstruction-based unsupervised drift detection over multivariate streaming data. In 2022 IEEE International Conference on Data Mining Workshops (ICDMW). doi:10.1109/ICDMW58026.2022.00109

Examples

#See an example of using `dfr_aedd` at this
#https://github.com/cefet-rj-dal/heimdall/blob/main/multivariate/dfr_aedd.md

Cumulative Sum for Concept Drift Detection (CUSUM) method

Description

CUSUM is a sequential analysis procedure that accumulates deviations in a monitored signal and raises an alarm when the cumulative evidence exceeds a threshold. In this package, the detector is implemented as an error-based monitor, so it is primarily intended for real concept drift affecting predictive performance. The concept-drift adaptation follows the sequential change-detection literature discussed by Muthukrishnan, Berg, and Wu (2007) doi:10.1109/ICDMW.2007.89.

Usage

dfr_cusum(lambda = 100)

Arguments

Value

dfr_cusum object

References

Muthukrishnan, S., Berg, E., and Wu, Y. (2007). Sequential change detection on data streams. In Seventh IEEE International Conference on Data Mining Workshops (ICDMW 2007). doi:10.1109/ICDMW.2007.89

Examples

library(daltoolbox)
library(heimdall)

# This example uses an error-based drift detector with a synthetic a 
# model residual where 1 is an error and 0 is a correct prediction.

data(st_drift_examples)
data <- st_drift_examples$univariate
data$event <- NULL
data$prediction <- st_drift_examples$univariate$serie > 4

model <- dfr_cusum()

detection <- NULL
output <- list(obj=model, drift=FALSE)
for (i in 1:length(data$prediction)){
 output <- update_state(output$obj, data$prediction[i])
 if (output$drift){
   type <- 'drift'
   output$obj <- reset_state(output$obj)
 }else{
   type <- ''
 }
 detection <- rbind(detection, data.frame(idx=i, event=output$drift, type=type))
}

detection[detection$type == 'drift',]

Adapted Drift Detection Method (DDM) method

Description

DDM monitors the online error rate of a predictive model under the PAC-learning assumption that, in a stationary environment, the error should decrease or remain stable as more samples are observed. Because it operates on the classifier error stream, it is primarily a detector of real concept drift. The method follows Gama et al. (2004) doi:10.1007/978-3-540-28645-5_29.

Usage

dfr_ddm(min_instances = 30, warning_level = 2, out_control_level = 3)

Arguments

Value

dfr_ddm object

References

Gama, J., Medas, P., Castillo, G., and Rodrigues, P. P. (2004). Learning with drift detection. In Advances in Artificial Intelligence - SBIA 2004, 286-295. doi:10.1007/978-3-540-28645-5_29

Examples

library(daltoolbox)
library(heimdall)

# This example uses an error-based drift detector with a synthetic a 
# model residual where 1 is an error and 0 is a correct prediction.

data(st_drift_examples)
data <- st_drift_examples$univariate
data$event <- NULL
data$prediction <- st_drift_examples$univariate$serie > 4

model <- dfr_ddm()

detection <- NULL
output <- list(obj=model, drift=FALSE)
for (i in 1:length(data$prediction)){
 output <- update_state(output$obj, data$prediction[i])
 if (output$drift){
   type <- 'drift'
   output$obj <- reset_state(output$obj)
 }else{
   type <- ''
 }
 detection <- rbind(detection, data.frame(idx=i, event=output$drift, type=type))
}

detection[detection$type == 'drift',]

Adapted EWMA for Concept Drift Detection (ECDD) method

Description

ECDD applies an exponentially weighted moving average (EWMA) control chart to the online classification error stream. Since it monitors predictive errors directly, it is primarily designed to detect real concept drift. The method follows Ross et al. (2012), who adapted EWMA charts for concept-drift detection in streaming classifiers doi:10.1016/j.patrec.2011.08.019.

Usage

dfr_ecdd(lambda = 0.2, min_run_instances = 30, average_run_length = 100)

Arguments

Value

dfr_ecdd object

References

Ross, G. J., Adams, N. M., Tasoulis, D. K., and Hand, D. J. (2012). Exponentially weighted moving average charts for detecting concept drift. Pattern Recognition Letters, 33(2), 191-198. doi:10.1016/j.patrec.2011.08.019

Examples

library(daltoolbox)
library(heimdall)

# This example uses an error-based drift detector where 1 is an error and 0 is a correct prediction.

data(st_drift_examples)
data <- st_drift_examples$univariate
data$event <- NULL
data$prediction <- st_drift_examples$univariate$serie > 4

model <- dfr_ecdd()

detection <- NULL
output <- list(obj=model, drift=FALSE)
for (i in 1:length(data$prediction)){
 output <- update_state(output$obj, data$prediction[i])
 if (output$drift){
   type <- 'drift'
   output$obj <- reset_state(output$obj)
 }else{
   type <- ''
 }
 detection <- rbind(detection, data.frame(idx=i, event=output$drift, type=type))
}

detection[detection$type == 'drift',]

Adapted Early Drift Detection Method (EDDM) method

Description

EDDM extends DDM by monitoring the distance between classification errors instead of only the error rate, which makes it more sensitive to gradual degradation. Because it operates on the model error stream, it is primarily intended for real concept drift. The method follows Baena-Garcia et al. (2006), who proposed EDDM for improved detection of gradual drift.

Usage

dfr_eddm(
  min_instances = 30,
  min_num_errors = 30,
  warning_level = 0.95,
  out_control_level = 0.9
)

Arguments

Value

dfr_eddm object

References

Baena-Garcia, M., del Campo-Avila, J., Fidalgo, R., Bifet, A., Gavaldà, R., and Morales-Bueno, R. (2006). Early drift detection method. In Fourth International Workshop on Knowledge Discovery from Data Streams.

Examples

library(daltoolbox)
library(heimdall)

# This example uses an error-based drift detector with a synthetic a 
# model residual where 1 is an error and 0 is a correct prediction.

data(st_drift_examples)
data <- st_drift_examples$univariate
data$event <- NULL
data$prediction <- st_drift_examples$univariate$serie > 4

model <- dfr_eddm()

detection <- NULL
output <- list(obj=model, drift=FALSE)
for (i in 1:length(data$prediction)){
 output <- update_state(output$obj, data$prediction[i])
 if (output$drift){
   type <- 'drift'
   output$obj <- reset_state(output$obj)
 }else{
   type <- ''
 }
 detection <- rbind(detection, data.frame(idx=i, event=output$drift, type=type))
}

detection[detection$type == 'drift',]

Adapted Hoeffding Drift Detection Method (HDDM) method

Description

HDDM_A is a sequential detector based on Hoeffding's inequality that tests whether the mean of the monitored error stream has increased beyond statistically expected fluctuations. Because this implementation is error-based, it is primarily targeted at real concept drift. The theoretical basis follows Frias-Blanco et al. (2015) doi:10.1109/TKDE.2014.2345382.

Usage

dfr_hddm(
  drift_confidence = 0.001,
  warning_confidence = 0.005,
  two_side_option = TRUE
)

Arguments

Value

dfr_hddm object

References

Frias-Blanco, I., del Campo-Avila, J., Ramos-Jimenez, G., Morales-Bueno, R., Ortiz-Diaz, A., and Caballero-Mota, Y. (2015). Online and nonparametric drift detection methods based on Hoeffding's bounds. IEEE Transactions on Knowledge and Data Engineering, 27(3), 810-823. doi:10.1109/TKDE.2014.2345382

Examples

library(daltoolbox)
library(heimdall)

# This example uses an error-based drift detector with a synthetic a 
# model residual where 1 is an error and 0 is a correct prediction.

data(st_drift_examples)
data <- st_drift_examples$univariate
data$event <- NULL
data$prediction <- st_drift_examples$univariate$serie > 4

model <- dfr_hddm()

detection <- NULL
output <- list(obj=model, drift=FALSE)
for (i in 1:length(data$prediction)){
 output <- update_state(output$obj, data$prediction[i])
 if (output$drift){
   type <- 'drift'
   output$obj <- reset_state(output$obj)
 }else{
   type <- ''
 }
 detection <- rbind(detection, data.frame(idx=i, event=output$drift, type=type))
}

detection[detection$type == 'drift',]

Inactive dummy detector

Description

Implements Inactive Dummy Detector

Usage

dfr_inactive()

Value

Drifter object

Examples

# See ?hcd_ddm for an example of DDM drift detector

KL Distance method

Description

This detector compares consecutive reference and recent windows through the Kullback-Leibler divergence estimated from their empirical distributions. In this package, it is primarily used for virtual concept drift, since it monitors changes in the distribution of a numeric feature stream rather than predictive error. The statistical foundation is the Kullback-Leibler divergence introduced by Kullback and Leibler (1951).

Usage

dfr_kldist(target_feat = NULL, window_size = 100, p_th = 0.05, data = NULL)

Arguments

Value

dfr_kldist object

References

Kullback, S., and Leibler, R. A. (1951). On information and sufficiency. The Annals of Mathematical Statistics, 22(1), 79-86. doi:10.1214/aoms/1177729694

Examples

library(daltoolbox)
library(heimdall)

# This example uses a dist-based drift detector with a synthetic dataset.

data(st_drift_examples)
data <- st_drift_examples$univariate
data$event <- NULL

model <- dfr_kldist(target_feat='serie')

detection <- NULL
output <- list(obj=model, drift=FALSE)
for (i in 1:length(data$serie)){
 output <- update_state(output$obj, data$serie[i])
 if (output$drift){
   type <- 'drift'
   output$obj <- reset_state(output$obj)
 }else{
   type <- ''
 }
 detection <- rbind(detection, data.frame(idx=i, event=output$drift, type=type))
}

detection[detection$type == 'drift',]

KSWIN method

Description

KSWIN applies a Kolmogorov-Smirnov test between a recent window and a reference sample drawn from older observations. In this package, the method is primarily used for virtual concept drift, because it monitors distributional changes in a numeric feature stream. The method follows Raab et al. (2020) doi:10.1016/j.neucom.2019.11.111.

Usage

dfr_kswin(
  target_feat = NULL,
  window_size = 1500,
  stat_size = 500,
  alpha = 1e-07,
  data = NULL
)

Arguments

Value

dfr_kswin object

References

Raab, C., Heusinger, M., and Schleif, F.-M. (2020). Reactive soft prototype computing for concept drift streams. Neurocomputing, 416, 340-351. doi:10.1016/j.neucom.2019.11.111

Examples

library(daltoolbox)
library(heimdall)

# This example uses a dist-based drift detector with a synthetic dataset.

data(st_drift_examples)
data <- st_drift_examples$univariate
data$event <- NULL

model <- dfr_kswin(target_feat='serie')

detection <- NULL
output <- list(obj=model, drift=FALSE)
for (i in 1:length(data$serie)){
 output <- update_state(output$obj, data$serie[i])
 if (output$drift){
   type <- 'drift'
   output$obj <- reset_state(output$obj)
 }else{
   type <- ''
 }
 detection <- rbind(detection, data.frame(idx=i, event=output$drift, type=type))
}

detection[detection$type == 'drift',]

Levene Based Drift Detection Method method

Description

LBDD is a window-based detector that compares the variability of reference and recent samples using Levene's test. Because it monitors changes in the distribution of an observed feature rather than model performance, it is primarily aimed at virtual concept drift. In this package, the detector follows the statistical-testing approach discussed by Giusti et al. (2021) for drift analysis, using Levene's variance test as its core mechanism.

Usage

dfr_lbdd(target_feat = NULL, alpha = 0.01, window_size = 1500)

Arguments

Value

dfr_lbdd object

References

Giusti, L., Carvalho, L., Gomes, A. T., Coutinho, R., Soares, J., and Ogasawara, E. (2021). Analysing flight delay under concept drift. Evolving Systems. doi:10.1007/s12530-021-09415-z

Examples

library(daltoolbox)
library(heimdall)

# This example uses a dist-based drift detector with a synthetic dataset.

data(st_drift_examples)
data <- st_drift_examples$univariate
data$event <- NULL

model <- dfr_lbdd(target_feat='depart_visibility')

detection <- NULL
output <- list(obj=model, drift=FALSE)
for (i in 1:length(data$serie)){
 output <- update_state(output$obj, data$serie[i])
 if (output$drift){
   type <- 'drift'
   output$obj <- reset_state(output$obj)
 }else{
   type <- ''
 }
 detection <- rbind(detection, data.frame(idx=i, event=output$drift, type=type))
}

detection[detection$type == 'drift',]

Mean Comparison Distance method

Description

MCDD is a window-based detector that compares the location of reference and recent samples by means of hypothesis tests on their central tendency. Because it monitors the distribution of observed features rather than predictive errors, it is primarily intended for virtual concept drift. In this package, the detector follows the statistical-testing perspective adopted by Giusti et al. (2021) for drift analysis.

Usage

dfr_mcdd(target_feat = NULL, alpha = 1e-08, window_size = 1500)

Arguments

Value

dfr_mcdd object

References

Giusti, L., Carvalho, L., Gomes, A. T., Coutinho, R., Soares, J., and Ogasawara, E. (2021). Analysing flight delay under concept drift. Evolving Systems. doi:10.1007/s12530-021-09415-z

Examples

library(daltoolbox)
library(heimdall)

# This example uses a dist-based drift detector with a synthetic dataset.

data(st_drift_examples)
data <- st_drift_examples$univariate
data$event <- NULL

model <- dfr_mcdd(target_feat='depart_visibility')

detection <- NULL
output <- list(obj=model, drift=FALSE)
for (i in 1:length(data$serie)){
 output <- update_state(output$obj, data$serie[i])
 if (output$drift){
   type <- 'drift'
   output$obj <- reset_state(output$obj)
 }else{
   type <- ''
 }
 detection <- rbind(detection, data.frame(idx=i, event=output$drift, type=type))
}

detection[detection$type == 'drift',]

Multi Criteria Drifter sub-class

Description

Implements Multi Criteria drift detectors

Usage

dfr_multi_criteria(drifter_list, combination = "or", fuzzy_window = 10)

Arguments

Value

Drifter object

Adapted Page Hinkley method

Description

The Page-Hinkley test is a sequential change-point detector that monitors cumulative deviations from a running mean and signals a change when those deviations grow persistently. In this package, the implementation is primarily used for virtual concept drift when it monitors a numeric feature stream, although the same statistic can also be applied to error streams to detect real concept drift. The method is based on Page (1954) and the later streaming adaptation popularized in data-stream mining.

Usage

dfr_page_hinkley(
  target_feat = NULL,
  min_instances = 30,
  delta = 0.005,
  threshold = 50,
  alpha = 1 - 1e-04
)

Arguments

Value

dfr_page_hinkley object

References

Page, E. S. (1954). Continuous inspection schemes. Biometrika, 41(1/2), 100-115. doi:10.2307/2333009

Examples

library(daltoolbox)
library(heimdall)

# This example assumes a model residual where 1 is an error and 0 is a correct prediction.

data(st_drift_examples)
data <- st_drift_examples$univariate
data$event <- NULL
data$prediction <- st_drift_examples$univariate$serie > 4


model <- dfr_page_hinkley(target_feat='serie')

detection <- c()
output <- list(obj=model, drift=FALSE)
for (i in 1:length(data$serie)){
 output <- update_state(output$obj, data$serie[i])
 if (output$drift){
   type <- 'drift'
   output$obj <- reset_state(output$obj)
 }else{
   type <- ''
 }
 detection <- rbind(detection, list(idx=i, event=output$drift, type=type))
}

detection <- as.data.frame(detection)
detection[detection$type == 'drift',]

Passive dummy detector

Description

Implements Passive Dummy Detector

Usage

dfr_passive()

Value

Drifter object

Examples

# See ?hcd_ddm for an example of DDM drift detector

Distribution Based Drifter sub-class

Description

Implements Distribution Based drift detectors

Usage

dist_based(target_feat)

Arguments

Value

Drifter object

Drifter

Description

Ancestor class for drift detection

Usage

drifter()

Value

Drifter object

Examples

# See ?dd_ddm for an example of DDM drift detector

Error Based Drifter sub-class

Description

Implements Error Based drift detectors

Usage

error_based()

Value

Drifter object

Examples

# See ?hcd_ddm for an example of DDM drift detector

Process Batch

Description

Process Batch

Usage

## S3 method for class 'drifter'
fit(obj, data, prediction, ...)

Arguments

Value

updated Drifter object

Metric

Description

Ancestor class for metric calculation

Usage

metric()

Value

Metric object

Examples

# See ?metric for an example of DDM drift detector

Accuracy Calculator

Description

Class for accuracy calculation

Usage

mt_accuracy()

Value

Metric object

Examples

# See ?mt_accuracy for an example of Accuracy Calculator

FScore Calculator

Description

Class for FScore calculation

Usage

mt_fscore(f = 1)

Arguments

Value

Metric object

Examples

# See ?mt_fscore for an example of FScore Calculator

Precision Calculator

Description

Class for precision calculation

Usage

mt_precision()

Value

Metric object

Examples

# See ?mt_precision for an example of Precision Calculator

Recall Calculator

Description

Class for recall calculation

Usage

mt_recall()

Value

Metric object

Examples

# See ?mt_recall for an example of Recall Calculator

ROC AUC Calculator

Description

Class for QOC AUC calculation

Usage

mt_rocauc()

Value

Metric object

Examples

# See ?mt_rocauc for an example of ROC AUC Calculator

Multivariate Distribution Based Drifter sub-class

Description

Implements Multivariate Distribution Based drift detectors

Usage

mv_dist_based()

Value

Drifter object

Norm

Description

Ancestor class for normalization techniques

Usage

norm(norm_class)

Arguments

Value

Norm object

Examples

# See ?norm for an example of DDM drift detector

Memory Normalizer

Description

Normalizer that has own memory

Usage

nrm_memory(norm_class = minmax())

Arguments

Value

Norm object

Examples

# See ?nrm_mimax for an example of Memory Normalizer

Reset State

Description

Reset Drifter State

Usage

reset_state(obj)

Arguments

Value

updated Drifter object

Examples

# See ?hcd_ddm for an example of DDM drift detector

Synthetic time series for concept drift detection

Description

A list of multivariate time series for drift detection

• example1: a bivariate dataset with one multivariate concept drift example

#'

Usage

data(st_drift_examples)

Format

A list of time series.

Source

Stealthy package

References

Stealthy package

Examples

data(st_drift_examples)
dataset <- st_drift_examples$example1

Stealthy

Description

Ancestor class for drift adaptive models

Usage

stealthy(
  model,
  drift_method,
  monitored_features = NULL,
  norm_class = daltoolbox::zscore(),
  warmup_size = 100,
  th = 0.5,
  target_uni_drifter = FALSE,
  incremental_memory = TRUE,
  verbose = FALSE,
  reporting = FALSE
)

Arguments

Value

Stealthy object

Examples

# See ?dd_ddm for an example of DDM drift detector

Update State

Description

Update Drifter State

Usage

update_state(obj, value)

Arguments

Value

updated Drifter object

Examples

# See ?hcd_ddm for an example of DDM drift detector