With the accelerating urbanization worldwide in 2026

urban transportation data exhibit the "4V" characteristics of volume

velocity

variety

and veracity

making traditional data processing architectures inadequate to meet the stringent real-time requirements of intelligent transportation systems. To address this challenge

this paper systematically reviews and analyzes real-time big data processing architectures for urban transportation based on Apache Spark. The study employs literature review

comparative analysis

and inductive methods

focusing on Spark's evolution from early batch processing to real-time stream processing via Structured Streaming. Results show that the "Kafka + Spark" decoupled architecture has become the de facto standard for handling high-concurrency traffic data streams

with Spark's in-memory computing mechanism and micro-batch processing model effectively reducing processing latency. Furthermore

the deep integration of Spark with MLlib demonstrates significant advantages in traffic flow prediction and congestion detection. The conclusion indicates that intelligent transportation data platforms built on Spark offer low latency

high throughput

and end-to-end exactly-once semantics

providing a solid theoretical foundation and practical engineering guidance for developing a new generation of reliable urban traffic perception and control systems.