Accurate aircraft arrival transit time predictions are critical for reliable, efficient airport traffic management. This task is made more challenging by the different airspace characteristics across airports. While recent data-driven models show promise, two key limitations remain, namely the exclusion of tactical arrival operations and inadequate weather consideration. In this study, we develop a two-stage gradient boosting framework for aircraft arrival transit time prediction, incorporating new weather and trajectory features. The framework decomposes the prediction problem into holding pattern classification and transit time regression, explicitly modeling operational decision-making processes. Specifically, we perform a case study on 58,378 arrival flights in 2018 at the Hong Kong International Airport (HKIA). We introduce several new features including Bayesian weather-induced traffic features, route-specific rainfall intensity metrics, and trajectory-based identifiers for Standard Terminal Arrival (STAR) assignments. Our results show that the proposed framework with these features significantly improves predictive accuracy, particularly under adverse weather conditions. The two-stage gradient-boosting framework achieves a 6.09 percentage point reduction in mean absolute percentage error (MAPE) under extreme weather scenarios. Our Bayesian weather-induced traffic features outperform the established ATMAP weather metric, demonstrating superior capability in capturing weather impacts on arrival times. This new framework provides a more comprehensive understanding of airspace characteristics. The use of data types that are commonly available in almost all airports in the feature derivation makes it possible to apply the same approach in other airports.