The risk of sophisticated cyberattacks against

real-time services has dramatically grown due to the quick

adoption of cloud computing in smart city infrastructures.

Due to their high false alarm rates and limited flexibility,

traditional security measures frequently fail to identify clever

and changing crimes. An intelligent hybrid learning approach

for real-time cloud security threat detection in smart cities is

proposed in this research. The suggested method combines

deep learning and machine learning approaches to improve

classification

accuracy,

identification,

temporal

attack

pattern

and feature representation. Standard

performance indicators, including as accuracy, precision,

recall, F1-score, false alarm rate, and detection time, are used

in extensive tests to assess the model's efficiency. According to

experimental data, the suggested hybrid model performs

better, achieving an accuracy of 96.3% and drastically

lowering the false alarm rate to 2.1%. Its viability for real

time deployment in cloud systems for smart cities is further

confirmed by the decreased detection time. The suggested

methodology provides a scalable and dependable way to

improve cloud security tolerance.