KEY TAKEAWAYS
- The study aimed to enhance SVM performance in cancer diagnosis by addressing high-dimensional gene expression data challenges, through IG-GPSO feature selection.
- Researchers noticed that the feature subset selected by IG-GPSO enhances classification accuracy and reduces FS.
Cancer diagnosis using machine learning has garnered significant attention, with Support Vector Machine (SVM) emerging as a widely adopted algorithm due to its efficacy with high-dimensional, small-sample data. However, the inherent high-dimensional feature space and redundancy in gene expression data pose challenges, leading to suboptimal classification outcomes for SVM.
Fangyuan Yang and the team aimed to address this by evaluating SVM’s performance in cancer diagnosis amidst such data challenges.
Researchers performed an inclusive analysis, proposing a hybrid feature selection algorithm termed information gain and grouping particle swarm optimization (IG-GPSO). The algorithm commences by computing information gain values for features and arranging them in descending order. Subsequently, features are grouped based on their information indices to ensure proximity within groups and sparsity outside. Lastly, a grouping Particle Swarm Optimization (PSO) method is employed to search the grouped features, evaluating their performance concerning to in-group and out-group.
The results and experimental findings demonstrate that the SVM achieves an average accuracy (ACC) of 98.50% on the feature subset selected by IG-GPSO, surpassing traditional feature selection methods significantly. Furthermore, compared to K-Nearest Neighbors (KNN), the classification performance of IG-GPSO’s selected feature subset remains optimal. Multiple comparison tests confirm the superior feature selection efficacy of IG-GPSO over conventional algorithms.
The study concluded that the feature subset chosen by IG-GPSO exhibits superior classification effectiveness and the least feature scale (FS) compared to alternatives. Most notably, IG-GPSO substantially enhances the ACC of SVM in cancer diagnosis.
The study was supported by the National Natural Science Foundation of China, the Nationally Funded Postdoctoral Researcher Program of China, the Key Science and Technology Program of Henan Province, and the Key Scientific Research Projects of Colleges and Universities in Henan Province.
Source: https://pubmed.ncbi.nlm.nih.gov/38466662/
Yang F, Xu Z, Wang H, et al. (2024). “A hybrid feature selection algorithm combining information gain and grouping particle swarm optimization for cancer diagnosis.” PLoS One. 2024 Mar 11;19(3):e0290332. doi: 10.1371/journal.pone.0290332. PMID: 38466662; PMCID: PMC10927139.
