Small Sample Scene Cancer Classification Method Based on Combined Feature Dimensionality Reduction

Accurately predicting cancer is a crucial task, especially in scenarios with small sample sizes. Machine learning-based cancer classification methods are widely employed due to their high accuracy. However, in small sample scenarios, the limited availability of training samples poses significant challenges. Therefore, this paper proposes a small-sample cancer classification method based on mixed feature dimensionality reduction. Firstly, it utilizes embedding-based feature selection technology to identify the most representative features. Secondly, the calculation of dynamic correlation weights is introduced to preserve the correlation between features. Finally, PCA technology is employed for feature extraction, and the optimal feature subset is extracted from small sample scenarios for model training. The model is then evaluated using common machine learning prediction models through in-depth comparative analysis. Numerous experiments demonstrate that the proposed model performs well in small sample scenarios. It effectively reduces the dimensionality of cancer data, accurately classifies different types of cancer, and achieves favorable prediction results.