EVALUATING HYBRID MODEL PERFORMANCE IN PREDICTING THE IMPACT OF LOCKDOWN ON CONTROLLING COVID-19 CASES PROGRESSION IN MIDDLE INCOME SETTINGS

Abstract

The rapidly spreading COVID-19 pandemic in 2020-2021 affected more than 190 countries, including Nigeria. Following this scenario, countries around the world were monitoring confirmed cases, recovering and dying. In order to reduce the impact of the pandemic, most countries implemented several measures to control the spread of the virus. These include closing schools and borders, shutting down public transport and workplaces, and restricting public gatherings until herd immunity is achieved through vaccination. The breakdown of the health system and the unpredictable nature of human behaviour makes it difficult to predict and evaluate the impact of lockdown on the COVID-19 pandemic in the long term. In light of the above, this study developed Hybrid ANN-CNN and four other models, namely LASSO, ANN, CNN and LSTM, to predict and evaluate the effect of human mobility on COVID-19 confirmed cases. To evaluate the effectiveness of non-pharmaceutical interventions (NPIs) and to predict the spread of COVID-19 confirmed cases, publicly available data on human mobility collected by Google and air passenger data were used. In this study, our motivation was to evaluate effect of lockdown on COVID-19 and models that predict the impact of human mobility on COVID-19 confirmed cases based on MSLE, Huber loss, and Log Cosh performance measures. At the end of the experiment, the developed hybrid ANN-CNN outperformed the other four models with MSLE of (0.0022), Huber Loss (0.0014) and Log Cosh (0.0013) respectively. This study serves as an alarm system to provide policy makers with the human mobility factors that can trigger large numbers of cases during a pandemic. This will allow for urgent public action.