Machine learning algorithms, according to research published in the International Journal of Data Science, may be used to make accurate predictions about population trends. According to the findings, the best available algorithms trained on historical data outperform traditional demographic modeling based on census data.
This study forecasted the population using various machine learning algorithms such as Extreme gradient boosting, CatBoost, linear and ridge regression, Holt-Winters, exponential, autoregressive integrated moving average (ARIMA), and prophet prediction model. Models were trained between 1960 and 2017 using 1595 different demographic variables from 262 different countries.
According to the findings, machine learning algorithms outperformed the demographic model. When the results of other algorithms were compared, the extreme gradient boosting model proved to be the most effective. Furthermore, the entire Turkish population in 2017 was approximated using pre-trained machine learning algorithms, and the results projected using the Cohort component technique were compared.
Fatih Veli ahinarslan, Ahmet Tezcan Tekin, and Ferhan ebi of Istanbul Technical University’s Department of Management Engineering in Istanbul, Turkey, compared extreme gradient boosting, CatBoost, linear regression, ridge regression, Holt-Winters, exponential, autoregressive integrated moving average (ARIMA), and prophet prediction model. To train the algorithms, they used 1595 different demographic statistics from 262 countries between 1960 and 2017. Age and gender distribution, labor force, education, birthplace, birth and death rates, and migration data are among the indicators.
In their presentation on predicting Turkey’s population for 2017, they demonstrated the advantage of the algorithmic technique over traditional modeling. Understanding population dynamics and forecasting how a population will change in the future is a critical component of healthcare, education, housing, transportation, and infrastructure policymaking and planning. Ten-year census cycles are useful, but they do not provide a fine-grained picture of a changing population, especially in light of changes in life expectancy, migration, war, political upheaval, and pandemics, all of which can cause the character of a population to change dramatically over a much shorter timeframe.
According to the researchers, machine learning techniques, specifically ensemble regression models, can provide a “better estimate” of a country’s future population. They may do so because they can reduce the number of components that would otherwise complicate the estimation and analyze any ambiguities in the demographic data.
The researchers concluded that machine learning algorithms for population prediction would be critical for national needs planning and would pave the way for more consistent social, economic, and environmental decisions.
