Online Change Detection for time-series

This is the implementation of the online time-series change detection algoirthm, a novel predictive model based method, that is more robust when the data are noisy and have outlier and runs in near-linear time. Two datasets are published in this repository:

• ** Enhanced Vegetation Index**: A remote sensing dataset that measures the “greenness” (area-averaged canopy photosynthetic capacity) as a proxy for the amount of vegetated biomass at a particular location [1]. We use our algorithm to autonomously detect forest fire from the dataset.
• ** Chestmounted Accelerometer data**: The original dataset is from the UCI Machine Learning Repository [2, 3]. We pre-process the data to encounter the impreciseness of the raw data.

Requirements

C++ 11, please update the compiler to be C++ 11 compiler.

Installation

Download the snapshot. To compile, simply run

make

Run Algorithm

usage: ./ts_cd data.txt data_path output_dir days tsLength n m w p st N

days: number of time series
tsLength: default time-series length
data_path: The path to the file containing the data, default format is txt
output_dir: The output directory for p-matrix and event matrix
n: train data length
m: repeat number to find median
w: window length
p: length of repeatable pattern
st: train length of event matrix
N: number of points in data

By default,

days=200
tsLength=285 
n=5 
m=50 
w=20 
p=23 
st=95 
N=10000

Reference:

[1] DAAC, LP. "Land processes distributed active archive center." (2012).
[2] Casale, Pierluigi, Oriol Pujol, and Petia Radeva. "Personalization and user verification in wearable systems using biometric walking patterns." Personal and Ubiquitous Computing 16.5 (2012): 563-580.
[3] Asuncion, Arthur, and David Newman. "UCI machine learning repository." (2007).