Logistics distribution in smart city construction has become the key to urban management. Traditional prediction models rely on historical data, but cannot quickly respond to emergencies such as traffic accidents and abnormal weather, resulting in poor accuracy and adaptability. This paper combines big data and machine learning to build a flexible prediction model to improve distribution efficiency and accuracy. In this model, first, ETL (Extract-Transform-Load) technology is used to collect and integrate data from multiple sources such as traffic flow, weather data, and historical orders to ensure high data quality and consistency. Real-time data updates are achieved through Internet of Things (IoT) sensors and Apache Kafka middleware to track the location and environmental changes of delivery vehicles in real time and quickly transmit data to the back-end system. Demand forecasting uses support vector machine (SVM) to train historical order data to predict future order demand. According to the prediction results, the A algorithm is used to optimize the delivery path, and the path is dynamically adjusted to adapt to real-time traffic and weather changes to ensure the minimization of delivery time and cost. The data shows that the prediction accuracy of the model is higher than 0.8, the recall rate is generally higher than 0.7, and the efficiency of path optimization is significantly improved.