Abstract: Prediction of micro-earthquakes when monitoring geothermal reservoirs, CO2 storage sites, and unconventional reservoirs using active source seismic data although challenging, is essential to ensuring the safety of the operations. Laboratory-scale friction experiments have shown that changes in elastic wave amplitude and speed during active source monitoring carry precursory information about the upcoming failure. The friction experiment is conducted close to the stability boundary producing numerous regular and irregular seismic cycles. A pair of the P-wave ultrasonic transducer is used to probe the laboratory fault throughout the seismic cycles. The data-driven deep learning models predict the timing and size (shear stress drop) of the laboratory earthquakes with great accuracy using the elastic wave attributes (physics-based wave speed and amplitude). In the active source ultrasonic monitoring, the ultrasonic signals need to be truncated to extract the wave features, and the rest of the information is discarded. A few hand-picked features are extracted from this reduced data. So, to overcome the data reduction limitation a novel method extracting the ultrasonic features automatically from the entire signal is demonstrated using a deep learning framework. In this, the recorded ultrasonic signals are passed through a convolutional neural network (CNN) to extract the features automatically. Furthermore, feature importance and an optimum number of feature selections are carried out using the XGBoost method. Finally, Long Short-term Memory (LSTM) model is developed using the time history of these features to predict the failure. Results show that the timing and size of the failure can be predicted with high confidence. The data-driven models require training on large datasets and overlook the physical laws controlling the shear failure. As such, the model may perform well for a particular dataset but fail to provide satisfactory predictions for a different albeit closely related dataset. To incorporate the domain knowledge and address the model transferability challenge, in this study, a physicsinformed deep learning approach is also implemented to forecast failure. The data-driven predictions obtained using the Multi-layer Perceptron (MLP) & LSTM model are used as a baseline. Along with shear stress, the shear failure rate (fault slip rate) is also predicted and used in the physical constraint formulations. The rate-and-state friction law and elastic coupling relation are integrated into the deep learning model architecture to modify the loss function. We hypothesize that a proposed physics-informed deep learning framework can improve model generalizability, prediction of an earthquake, and micro-seismicity informed by the physics laws dictating the shear failure state.