Title: Retiring in the informal economy: Implications for social policyintervention for ageing workers in Ghana

Abstract

The COVID-19 pandemic, starting in December 2019, has ravaged the whole world. Scholars from all over the world have utilized traditional and advanced approaches to detect and forecast the spreading of this disease. However, most of these researches need a large number of features and private information which are not easily accessed. Moreover, single models cannot achieve satisfactory results. To address the above two issues, we propose a hybrid machine learning method. The opensource dataset we use stems from a Kaggle competition: “COVID19 Global Forecasting (Week 4)”, which collected the confirmed cases and fatalities as well as the date and region information provided by the JHU CSSE (Johns Hopkins University Center for Systems Science and Engineering). After finishing preliminary feature engineering, we construct a hybrid machine learning model using stack techniques and integrate LASSO (Least Absolute Shrinkage and Selection Operator), SVM (Support Vector Machine), RF(Random Forest), LGBM(Light Gradient Boosting Machine), and Linear Regression models to fit and forecast the transmission of COVID-19. Despite of limited features, the model obtains 0.5350 as its coefficient of determination and outperforms the baselines. We plot the model in Figure 1. Model RMSE of Confirmed Cases RMSE of Fatalities Ours 13041.93 1348.30 SIR 160367.25 368640.15 Table 1 shows that our proposed hybrid model surpasses the SIR model in terms of the evaluation metric: RMSE (Root Mean Square Error).In conclusion, our work shows satisfactory performance in COVID-19 forecasting and provides promising application prospects for combining medicine and artificial intelligence.