Disclosed are a system for predicting optical properties of molecules based on machine learning and a method thereof. More particularly, the system for predicting optical properties according to an embodiment includes a preprocessor that receives molecular information of a target molecule and surrounding molecules, and vectorizes the molecular information of a target molecule and surrounding molecules; a feature extractor that receives the vectorized information of the target molecule and surrounding molecules and extracts the features of the target molecule and surrounding molecules; an integrated feature extractor that receives both features of the target molecule and surrounding molecules and extract the integrated features of the target molecule and surrounding molecules by using an algorithm; and an optical property predictor that receives the integrated features of the target molecule and surrounding molecules and predicts optical properties of the target molecule affected by surrounding molecules.