In this paper, we propose a novel method named CNN-AE to predict survival chance of COVID-19 patients using a CNN trained on clinical information. To further increase the prediction accuracy, we use the CNN in combination with an autoencoder. Our method is one of the first that aims to predict survival chance of already infected patients. We rely on clinical data to carry out the prediction. The motivation is that the required resources to prepare CT images are expensive and limited compared to the resources required to collect clinical data such as blood pressure, liver disease, etc. We evaluate our method on a publicly available clinical dataset of deceased and recovered patients which we have collected. Careful analysis of the dataset properties is also presented which consists of important features extraction and correlation computation between features. Since most of COVID-19 patients are usually recovered, the number of deceased samples of our dataset is low leading to data imbalance. To remedy this issue, a data augmentation procedure based on autoencoders is proposed. To demonstrate the generality of our augmentation method, we train random forest and Na\"ive Bayes on our dataset with and without augmentation and compare their performance. We also evaluate our method on another dataset for further generality verification. Experimental results reveal the superiority of CNN-AE method compared to the standard CNN as well as other methods such as random forest and Na\"ive Bayes. COVID-19 detection average accuracy of CNN-AE is 96.05% which is higher than CNN average accuracy of 92.49%. To show that clinical data can be used as a reliable dataset for COVID-19 survival chance prediction, CNN-AE is compared with a standard CNN which is trained on CT images.