Background: To assist policy makers in taking adequate decisions to stop the\nspread of COVID-19 pandemic, accurate forecasting of the disease propagation is\nof paramount importance. Materials and Methods: This paper presents a deep\nlearning approach to forecast the cumulative number of COVID-19 cases using\nBidirectional Long Short-Term Memory (Bi-LSTM) network applied to multivariate\ntime series. Unlike other forecasting techniques, our proposed approach first\ngroups the countries having similar demographic and socioeconomic aspects and\nhealth sector indicators using K-Means clustering algorithm. The cumulative\ncases data for each clustered countries enriched with data related to the\nlockdown measures are fed to the Bidirectional LSTM to train the forecasting\nmodel. Results: We validate the effectiveness of the proposed approach by\nstudying the disease outbreak in Qatar. Quantitative evaluation, using multiple\nevaluation metrics, shows that the proposed technique outperforms state-of-art\nforecasting approaches. Conclusion: Using data of multiple countries in\naddition to lockdown measures improve accuracy of the forecast of daily\ncumulative COVID-19 cases.\n