Design of Intelligent Decision-Making Model for Big Data Analysis

The existing decision-making system has problems such as low efficiency of multi-source heterogeneous data fusion and insufficient real-time decision-making accuracy. This paper improves the performance of the decision model by constructing a hybrid framework of fusion data preprocessing, ensemble learning and deep reinforcement learning. First, the distributed data cleaning technology based on MapReduce is used to process TB-level industrial sensor data to remove outliers and redundant information. Then, a dual-channel feature extraction structure is designed on the e-commerce user behavior dataset to compress the feature representation dimension to 23% of the original data. Subsequently, an improved deep deterministic policy gradient algorithm is introduced to optimize the decision strategy through a dynamic reward function, and a fraud recognition accuracy rate of 92.5% is achieved in a financial risk control simulation experiment. The online decision-making system finally deployed can process an average of 15,628 real-time data per second in the logistics scheduling scenario, and the average decision delay is controlled at 46.5ms, which verifies that the model improves data processing efficiency in complex scenarios and provides a scalable technical framework for cross-domain intelligent decision-making systems.