Predicting earthquakes remains a significant challenge in seismology due to the complicated nature of seismic events. Recent advancements in ML offer promising solutions to this problem by utilizing large datasets and advanced algorithms to identify patterns in seismic data. A real earthquake dataset from Kyrgyzstan was analyzed using eight different models: LR, ANNs, SVR, RF, GBM, LSTM, LightGBM, and XGBoost—to predict earthquake magnitudes. The evaluation of the models involved performance metrics like R-squared (R2), MSE, RMSE, and MAE. The key finding of this study is that LR provided the best fit for the dataset, achieving the highest R2 value among the evaluated models. This research helps in selecting suitable ML techniques for earthquake prediction, showing that both simple and complex models can be useful. It also highlights how ML could improve earthquake forecasting, especially in real-time situations.