Accurately assessing the depth of anesthesia (DoA) is vital for patient safety during surgery. Traditional methods rely on continuous monitoring and the anesthesiologist’s judgment, which can be resource-intensive. In this paper, we employ a machine learning model, LightGBM, to predict the DoA using features extracted from electroencephalogram (EEG) signals into distinct classes representing of anesthetic depth. To enhance interpretability, we employ SHapley Additive exPlanations (SHAP), which assigns importance values to each feature, ensuring transparency in our model’s predictions. Using LightGBM, we achieved an accuracy of 96.85% using 318 features extracted from the EEG signals and 96.34% on a smaller feature set selected via SHAP, demonstrating that even with fewer features, the model maintains high performance. The proposed scheme offers comparable or superior performance to existing methods while being more efficient and explainable. This cost-effective solution provides an accurate and transparent tool for determining DoA, paving the way for more standardized and objective anesthesia monitoring practices.