A Machine learning model for human population forecasting: case for Kenya

Abstract

The growth of a country’s population can be a complex issue that has a significant impact on the development and sustainability of countries all over the world. In Kenya, the population is growing rapidly, which is putting a strain on the resources of the country, such as land, water, and infrastructure. The currently used methods of forecasting population growth, such as censuses and mathematical models, are costly, time-consuming, and not consistently accurate. The aim of this study is to develop a ML algorithm to forecast population growth in Kenya more accurately compared to the models currently being used. In this study, seven different machine learning models were examined Artificial Neural Networks, Random Forest, Logistic Regression, Support Vector Machines, Linear Regression, Decision Trees, and K-Nearest Neighbor to determine their effectiveness in predicting the population of Kenya. A variety of factors that impact population growth were considered, such as fertility and mortality rates, life expectancy, net migration, economic growth, access to healthcare and education, and gender equality. All models were built using the Scikit-Learn library and demonstrated impressive accuracy, but the top performers were Artificial Neural Networks, Random Forest, and Linear Regression. Of these, Linear Regression stands out as the best performer overall with a MAPE of 0.0179% and an accuracy of 0.9977% when tested with new data. This is a significant improvement over the other models, which showed slightly lower accuracies.