Efficient representation of boolean decision structures through Boolean function optimization

A binary decision tree (BDT) is stochastic and depth-dependent when inference is performed. The lower and upper bounds are derived from the minimum and maximum heights of the leaf nodes. The inherent randomness complicates BDT and random forest (RF) inference processes for fixed-rate streaming data. BDT is reformulated as a Boolean decision structure (BDS) in the proposed method to enable constant-time complexity. Optimized BDS (OBDS) is constructed by aggregating decision nodes exhibiting approximate boundary similarity. Further optimization of the Boolean function is achieved by applying the ESPRESSO algorithm on OBDS (EOBDS). Based on empirical evidence, BDS, OBDS, and EOBDS are statistically equivalent on BDT and BDT-based RF, and exhibit a constant time complexity of inference, regardless of BDT depth or number of BDTs.