There are increasingly more discussions on and guidelines about different levels of indicators surrounding smart cities. These indicators might help effectively provide quality services to citizens because smart cities involve not only technical elements but also complex elements (e.g., comfort, well-being and weather conditions). They are an important opportunity to illustrate how smart urban development strategies and digital tools can be stretched or reinvented to address localised social issues. Thus, multi-source heterogeneous data provide a new driving force for exploring urban human mobility patterns. In this survey, we forecast human mobility data using LinkNYC kiosks and MTA Wi-Fi Locations in New York City to provide a large-scale study on how people's habits influence comfort and well-being indicators. By comparing the forecasting performance of statistical and deep learning methods on the aggregated mobile data we show that each class of methods has its advantages and disadvantages depending on the forecasting scenario. However, for our time-series forecasting problem, deep learning methods are preferable when it comes to simplicity and immediacy of use, since they do not require a time-consuming model selection for each different cell. Deep learning approaches are also appropriate when aiming to reduce the maximum forecasting error. Statistical methods instead have shown their superiority in providing more precise forecasting results, but they require data domain knowledge and computationally expensive techniques in order to select the best parameters.