Penerapan ML dengan Teknik Bayesian Regulation untuk Peramalan Usia Penduduk di Beberapa Negara Asia

Abstract

Knowing the age of life of the population in a country is useful for evaluating the performance of the government, whether the government is able to prosper the population in general, and improve health status in particular. The purpose of this paper is to forecast the age of the population in several major countries in Asia, so that the government has a benchmark in determining policies to further improve the welfare and health of the population in their respective countries. The forecasting method in this paper will use Machine learning algorithms with Bayesian Regulation techniques. The research data used is data on population expectations in several Asian countries sourced from the United Nations: "World Population Prospect: The 2010 Revision Population Database". This research is a development of research that has been done before, using the Cyclical order technique. Previous research used 5 architectural models (3-5-1, 3-8-1, 3-10-1, 3-5-8-1 and 3-5-10-1), with the best model being 3-5-10 -1 which results in an accuracy of 97%, MSE 0.0008358919, training time of 27 seconds and an error rate of 0.03. Meanwhile, this research only uses 3 modified architectural models (2-5-1, 2-10-1 and 2-5-10-1), with the best model being 2-5-1. The result is that this study is better than previous studies. The benchmark is seen from a smaller error rate (0.02), better accuracy (100%), to a faster training time (5 seconds). So it can be concluded, Bayesian Regulation technique works better than Cyclical order and the 2-5-1 architectural model will be used to make predictions