Feature Selection for Computer-Aided Diagnosis via a Novelty Designed Binary Harris Hawk Algorithm

Computer-aided diagnosis (CAD) has become one of the hot research directions in the combination of medicine and machine learning (ML). While the characteristic of high feature number and low sample size in disease datasets makes it difficult for ML to make effective predictions. Currently, one of the most commonly used methods to address this issue is the use of heuristic algorithms for feature selection (FS). However, traditional heuristic algorithms still have certain limitations. On one hand, single-objective algorithms only focus on model's accu-racy, which always leads to a large number of redundant features remaining after FS. On the other hand, while multi-objective algorithms may result in fewer features, they often significantly reduce accuracy compared to single-objective algorithms. To solve these shortcoming, we proposes a novel FS algorithm based on harris hawk optimization (HHO) algorithm, attempting to achieve a better balance between the number of features and model's accuracy. Specifically, since the original HHO algorithm was only used for continuous numerical problems, we defines a new set of discrete operations for HHO to make it suitable for feature selection. Additionally, we introduce an aggregation function that allows the model to consider both model's accuracy and the number of features with a single objective. To mitigate the accuracy degradation caused by the aggregation function, a scale-free network was employed to improve the model's optimization capability. To comprehensively evaluate the proposed algorithm, we compare it with commonly used single-objective algorithms and multi-objective algorithms. Experimental results show that the proposed algorithm offers a better balance compared to commonly used heuristic algorithms.