One of the challenges of living in today's cities is parking availability. Searching for available parking spots can be a time-consuming task that simultaneously increases traffic congestion and greenhouse gas pollution by a significant 40%. A solution that would increase drivers' ability to locate an empty parking spot would represent an important step towards more sustainable parking as it would have a direct impact on reducing greenhouse gas pollution in urban areas. This paper proposes how data science can help by evaluating the prediction performance of four machine learning models. Analysed machine learning models are based on different machine learning methods (i.e., CatBoost and Random Forest) and use different real-world data sets (i.e., parking sensor data only or contextually enriched parking sensor data). The dummy (baseline) model is considered as well, but with a R2 score of 61.29% is outperformed by more advanced data science approaches. Prediction performance in the case of using parking sensor data only gives R2 score of 84.31% and 88.16% for Random Forest and CatBoost, respectively. The best prediction performance is achieved using CatBoost and contextually enriched data, resulting in the high-performing machine learning model with the R2 value of 88.83%, thus outperforming the Random Forest model by 1.7%. In fact, for both machine learning methods, the contextually enriched data approach gives better results for predicting parking spot availability. This suggests that parking data should be enriched with contextual data when designing and building sustainable parking solutions for smart cities of the future.