Recent studies try to use hyperspectral imaging (HSI) to detect foreign matters in products because it enables to visualize the invisible wavelengths including ultraviolet and infrared. Considering the enormous image channels of the HSI, several dimension reduction methods-e.g., PCA or UMAP-can be considered to reduce but those cannot ease the fundamental limitations, as follows: (1) latency of HSI capturing. (2) less explanation ability of the important channels. In this paper, to circumvent the aforementioned methods, one of the ways to channel reduction, on anomaly detection proposed HSI. Different from feature extraction methods (i.e., PCA or UMAP), feature selection can sort the feature by impact and show better explainability so we might redesign the task-optimized and cost-effective spectroscopic camera. Via the extensive experiment results with synthesized MVTec AD dataset, we confirm that the feature selection method shows 6.90x faster at the inference phase compared with feature extraction-based approaches while preserving anomaly detection performance. Ultimately, we conclude the advantage of feature selection which is effective yet fast.