Small-Scale Demographic Sequences Projection Based on Time Series Clustering and LSTM-RNN

Abstract

Neural Networks are one of many data mining analytical tools that can be utilized to make predictions for demographic sequences. In this paper, we presented the application of a hybrid model that integrates Long-Short Term Memory-Recurrent Neural Network (LSTM-RNN), time series analysis and clustering techniques where time series analysis and clustering methods provide augmentation of sequences data for the training of RNN. Comprehensive characteristics of nations from UN database are used as input to the hybrid model to predict the nations future population. The results prove that, RNN combined with time series and clustering methods has outperformed mere RNN approach without time series and clustering analysis. In addition, the hybrid Time Series and Clustering-RNN with relevant inputs lead to 20% higher predictive accuracy, measured by Root-Mean-Squared Error, compared to results produced by RNN alone.