AnomalyCleaner

This is an R script that cleans anomalies. It is based on Twitter's AnomalyDetection 📦

Description

A technique for detecting anomalies in seasonal univariate time series. The methods uses are robust, from a statistical standpoint, in the presence of seasonality and an underlying trend.
For more information, you can review the paper Hochenbaum, J., Vallis, O. S., & Kejariwal, A. (2017). Automatic anomaly detection in the cloud via statistical learning. arXiv preprint arXiv:1704.07706.
Another important thing is that I've used a forked version of the original library which is maintained by (@hrbrmstr) and you can check it here

Required libraries

First you have to install the properly AnomalyDetection library using this:

install.packages("devtools")
devtools::install_github("hrbrmstr/AnomalyDetection")

Then the libraries that are used in this script are:

library(lubridate)
library(AnomalyDetection)
library(hrbrthemes)
library(tidyverse)
library(ggplot2)

Running Example

You can run the script using RStudio. This uses the data from Test_Git.csv which is a simple borehole level data set. This type of data usually has many anomalies.
After it runs, the script will create 2 files:

• Anom_plots
  Where you will have all the plots of your data and its anomalies, just like this:

• Filter_Output
  Where you will have your dataset cleaned.

Data's Format

If you want to use your own data you will have to make this kind of .csv files:

Name1	Tiempo1	Nivel1
Well1	13/10/1982	683.338
Well1	14/10/1982	683.942
Well1	15/10/1982	682.167
...	...	...
Well7	09/12/1983	694.684
Well7	10/12/1983	694.618
Well7	11/12/1983	694.255

Where Name1 is the ID of the dataset. Tiempo1 is the date of the data and Nivel1 is the observation. It is important to maintain this columns names while using this version.

Important

This is just the first version, I want to add more functionalities like smoothing the output data.

Contact

If you want to improve the code, please feel free to contact me or just make a pull request.
email: sebastian.vazquez@ug.uchile.cl
Linkedin