TY - GEN

T1 - Predicting Poverty Percentage Based on Satellite Imagery and Point of Interest Using Support Vector Regression and Random Forest Regression (Case Study of Central Java Province)

AU - Putra, I. Komang Pande Prajadhita Wibawa

AU - Irhamah,

AU - Iriawan, Nur

AU - Fithriasari, Kartika

N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024.

PY - 2024

Y1 - 2024

N2 - The percentage of poverty in Indonesia is quite high, on the island of Java there are provinces that have a percentage above 10%, one of which is the province of Central Java. This problem occurs because data collection is carried out conventionally with surveys and censuses so that it requires human resources, time, and large costs. Remote sensing that uses satellite imagery and Point of Interest (POI) data can provide lower costs and shorter time. The use of machine learning is often used in predicting poverty by using Support Vector Regression (SVR) with Random Forest Regression (RFR). Satellite image and POI were extracted using zonal statistics, consisting of NTL, NDVI, NDBI, NDWI, LST, CO, SO2, NO2, and POI density data. Estimating the poverty percentage in Central Java in 2021 using a method between SVR and RFR with a tenfold cross validation procedure. The regions in Central Java with low poverty percentages are Semarang City and Salatiga. There are 12 districts that have a low poverty percentage. The best model to estimate poverty in Central Java is the SVR model with the lowest MAPE, MAE, and RMSE values. The prediction results of poverty percentage in Central Java get 20 districts correctly predicted. The correlation value between actual and predicted is quite high and the average percentage error value is quite low so the model obtained is optimal.

AB - The percentage of poverty in Indonesia is quite high, on the island of Java there are provinces that have a percentage above 10%, one of which is the province of Central Java. This problem occurs because data collection is carried out conventionally with surveys and censuses so that it requires human resources, time, and large costs. Remote sensing that uses satellite imagery and Point of Interest (POI) data can provide lower costs and shorter time. The use of machine learning is often used in predicting poverty by using Support Vector Regression (SVR) with Random Forest Regression (RFR). Satellite image and POI were extracted using zonal statistics, consisting of NTL, NDVI, NDBI, NDWI, LST, CO, SO2, NO2, and POI density data. Estimating the poverty percentage in Central Java in 2021 using a method between SVR and RFR with a tenfold cross validation procedure. The regions in Central Java with low poverty percentages are Semarang City and Salatiga. There are 12 districts that have a low poverty percentage. The best model to estimate poverty in Central Java is the SVR model with the lowest MAPE, MAE, and RMSE values. The prediction results of poverty percentage in Central Java get 20 districts correctly predicted. The correlation value between actual and predicted is quite high and the average percentage error value is quite low so the model obtained is optimal.

KW - Machine learning

KW - Poverty

KW - Satellite imagery

UR - http://www.scopus.com/inward/record.url?scp=85200683929&partnerID=8YFLogxK

U2 - 10.1007/978-981-97-2136-8_23

DO - 10.1007/978-981-97-2136-8_23

M3 - Conference contribution

AN - SCOPUS:85200683929

SN - 9789819721351

T3 - Springer Proceedings in Mathematics and Statistics

SP - 309

EP - 323

BT - Applied and Computational Mathematics - ICoMPAC 2023

A2 - Adzkiya, Dieky

A2 - Fahim, Kistosil

PB - Springer

T2 - 8th International Conference on Mathematics: Pure, Applied and Computation, ICoMPAC 2023

Y2 - 30 September 2023 through 30 September 2023

ER -