Earthquake prediction is a complex task as it depends on many parameters such as complex spatio-temporal factors involving various components like complex signal structures, geology, earthquake pattern history, tectonic dynamics, and stress accumulation. Studying earthquake hazards and risks is crucial for predicting earthquake magnitude and depth for early warning systems, as it helps calculate seismic hazard analysis. Traditional statistical methods may be too inadequate to generalize a stochastic phenomenon such as earthquakes. However, artificial neural network architectures with stochastic structures such as deep neural networks will be more advantageous in terms of prediction in these cases. This study proposes Convolutional Long Short-Term Memory (ConvLSTM) for predicting earthquake parameters of magnitude and depth. It integrates convolutional neural networks (CNN) for spatial feature extraction and Long Short-Term Memory (LSTM) networks for temporal pattern learning, making it ideal for seismic data analysis due to its ability to handle time series and specific geology. In this study, with requested AFAD API data, ConvLSTM models are trained, including spatial coordinates, time series data, magnitude, and depth information. By leveraging ConvLSTM, we aim to improve earthquake parameter estimation and provide a data-driven approach to enhance seismic hazard assessment and early warning systems. Preliminary results demonstrate the effectiveness of ConvLSTM in capturing earthquake patterns and accurately predicting seismic event characteristics.