Flight arrival delay is a critical operational challenge that affects passenger experience, airline efficiency, and airport resource planning in the aviation industry. This study develops a machine learning-based prediction model for flight arrival delays at King Khalid International Airport in Riyadh, Saudi Arabia, by combining historical flight data with meteorological observations. A dataset of 47,107 flight arrival records collected from Flightera covering the period June 2025–September 2025 was integrated with hourly weather data for the same period obtained from Visual Crossing. The integrated dataset underwent systematic cleaning, preprocessing, and feature engineering, resulting in a final dataset of 42,361 records and 33 features. Three gradient boosting regression models, CatBoost, XGBoost, and LightGBM, were trained and evaluated using an 80:20 train-test split. The model outputs were converted into binary delay classifications using a 15-min threshold, and the performance was assessed using accuracy, precision, recall, and F1-score. LightGBM achieved the best overall performance, recording an accuracy of 94.45%, precision of 77.40%, and F1-score of 60.70%. Feature importance analysis revealed that departure delay, origin airport, and airline identity were the strongest predictors, while weather-related variables such as wind gust and sea-level pressure also contributed meaningfully. These findings indicate that integrating operational flight data with hourly weather information can support more accurate arrival-delay prediction and assist airport operators and airlines in planning resources, managing disruptions, and improving operational decision-making.