| Incremental Interval Regression Tree Learning with Mean Variance Numerical Data Streams | Inkrementalno izucavanje intervala primenom drveta regresije sa srednjom varijansom numerickih tokova podataka |
| Management, [pdf] | Menadžment, [pdf] |
| ID: 7.2012.63.3 Number: 63 Year: 2012 UDC: 005.521 ; 005.82 [tmx] [bow] |
Dima Alberg
Institution: Department of Industrial Engineering and Management, SCE - Shamoon College of Engineering, Beer-Sheva, Israel | Dima Alberg
Institucija: Odsek za industrijski inženjering i upravljanje, SCE - Shamoon College of Engineering, Beer-Sheva, Israel |
Abstract
In this paper, we present a novel method for interval regression tree incremental learning with mean variance
patterned numerical data streams. The proposed Mean Variance Interval Regression Tree (MVIRT) algorithm
transforms continuous temporal data into two statistical moments according to a user-specified time resolution
and builds a regression model tree for estimating the prediction interval of the target variable. The algorithm
main properties are time - based incremental mean variance tree induction algorithm accompanying
novel time resolution and outliers detection mechanism. Results of real world data stream show that the MVIRT
algorithm produces more accurate and easily interpretable prediction models than other state-of-the-art batch
incremental model tree methods.
| Apstrakt
U ovom radu predstavljamo novi model inkrementalnog ucenja primenom drveta regresije koristeci numericke tokove podataka struktuirane pomocu srednje varijanse. Predloženi MVIRT (Mean Variance Interval Regression Tree) algoritam pretvara kontinuirane vremenske podatke u dva statisticka momenta u skladu sa vremenom koje je korisnik odredio i gradi model stabla regresije za procenu intervala predvidljivosti ciljne varijable. Glavna osobenost ovog algoritma jeste vremenski odreden algoritam za indukciju inkrementalne varijanse koji se kombinuje sa novom rezolucijom vremena i mehanizmom za detekciju podataka koji odstupaju od uobicajenih. Rezultati tokova podataka u realnom vremenu pokazuju da se primenom MVIRT algoritma dobijaju precizniji modeli predvidanja koje je lakše tumaciti u poredenju sa drugim metodamama za serijsku obradu inkrementalnog modela drveta koji su danas u upotrebi.
|
| Keywords: prediction, regression tree, incremental learning, data stream mining, interval prediction | Ključne reči: predviđanje, drvo regresione analize, inkrementalno učenje, analiza tokova podataka, predviđanje intervala |
| Pages: 27-33 | Strane: 27-33 |