1PhD, Center for Technology and Systems Management, 4298 Campus Drive, Martin Hall, University of Maryland, College Park, MD, scmishra@t.edu
2PhD, Professor and Director, Center for Technology and Systems Management, 4298 Campus Drive, Martin Hall, University of Maryland, College Park, MD, ba@umd.edu
Online published on 13 June, 2019.
This study presents a novel deep learning framework where Dynamic Time Warping (DTW), Hierarchical Clustering Analysis (HCA) and long-short term memory (LSTM) are used to predict annual state level per capita Net GDP (NSDP) growth using data from the Reserve Bank of India (RBI) and Niti Ayog. The proposed ensemble model framework consists of the following: DTW and HCA are first used to identify similar states based on per capita NSDP growth trends, and similar socio-economic-demographic features within a given state, to construct a fine-tuned training-data set to predict each state's NSDP per capita growth. The saved training data is fed into a deep LSTM neural network for time-series predictions for NSDP per capita growth. Finally, MC dropout technique is used to quantify epistemic uncertainty to go beyond just the prediction point estimate. The developed model can be used to inform policy-makers using a highly accurate method for uncertain-adjusted prediction for regional economic activity.
Deep neural networks, NSDP prediction, Regional economic activity, Epistemic uncertainty, Economic development
