A Design for AI Machine Learning to Predict Economic Recession in Indonesia
Umar Tsani Abdurrahman, Wilarso

Sekolah Tinggi Teknologi Muhammadiyah Cileungsi

Abstract

A recent rapid change in World Economic landscape due to Covid-19 pandemic has been given a wake-u call for economists and scientists to be able to predict an upcoming economic recession. Since country Gross Domestic Product (GDP) including Indonesia is measured quarterly then it is required that a team of professionals or a system be able to predict it in upcoming quarter economic performances. Unlike in developed countries especially in US, there is a proven forum called Survey of Professional Forecaster there which has provided a reliable quarterly economic forecast for the country for decades. Indonesia is still lack of dedicated professional to do such job except for a National Survey Agency (BPS) which is a governmental body which its job is to provide economic quarterly statistics and performances for the National government. There is only a handful of economists and university scholars who usually have their opinion and forecast for the late country economic progress. So it will be helpful if we can provide a second opinion using AI Technology especially machine learning (ML) to provide a simple prediction at first. Since this is our first attempt to develop macroeconomic predictors using Machine Learning, we are simplifying the economic parameters to only include GDP and Oil Prices since the data is readily available. Based on previous study for forecasting the American economic recession, we chose the same Machine Learning algorithm known as Random Forrest. And to simplify design implementation we are using latest open source technology in machine learning, TensorFlow 2.0 (TF 2.0). Since its first release TF 1.0 in 2016 TensorFlow has become popular ML engine and as per last update the random Forrest algorithm has been implemented on 2.0. The Python Economic Prediction scripts for TF2.0 implementation we developed are provide here. Since this is considered not a computationally intensive configuration due to the simplified parameters used, we are not using

Keywords: Tensorflow, Random Forest, Indonesia GDP, Economic, Forecast

Topic: Computer Science