Forecasting migration intention using multivariate time series

• This paper aims to analyze international migrations in western African countries using irregular multivariate monthly time series containing a few values. Existing methods of filling in missing values have limitations because there are not enough values to infer them. In this study, we explore two approaches to solve this problem. One approach is to aggregate the values annually to eliminate missing values. The other is to use the Random Forest (RF) based approach to fill in the missing values. Then, we predict the international migration intentions using deep learning approaches and time series dataset. We demonstrate that a RF-based imputation outperforms a zero filling approach (used as the baseline) with Long Short-Term Memory (LSTM) method. Moreover, we show that analyzing the monthly subregion-based time series provides better insights than the yearly country-based time series. 

Subject headings

NATURVETENSKAP  -- Data- och informationsvetenskap (hsv//swe)

NATURAL SCIENCES  -- Computer and Information Sciences (hsv//eng)

NATURVETENSKAP  -- Data- och informationsvetenskap -- Datavetenskap (hsv//swe)

NATURAL SCIENCES  -- Computer and Information Sciences -- Computer Sciences (hsv//eng)

Keyword

Forecasting

Imputation

Long Short-term Memory

Migration Intention

Recurrent Neural Network

Decision trees

Deep learning

Image processing

Regional planning

Time series

Filling in

Learning approach

Missing values

Multivariate time series

Zero filling

Skövde Artificial Intelligence Lab (SAIL)

Skövde Artificial Intelligence Lab (SAIL)

Publication and Content Type

ref (subject category)

kon (subject category)