What data do you have available for training and testing?

I think, first, the enough training data is necessary. Second, you should find an effective learning model, such as CNN for images informations and DBN for time-series predictions. Most importantly, the learning model you used should has capability of avoiding local minimum. You can try the adapative learning rate and initilizing the weight paprameters using unsupervised manner based on training data.

Thanks.

Article Predicting COVID-19 in China Using Hybrid AI Model

Article Early Prediction of the 2019 Novel Coronavirus Outbreak in t...