Table 4 Most influential visual and statistical features used by the proposed model for insect classification, as jointly identified by SHAP and PFI (top-ranked subset).

1

hog_1757

Visual (HOG gradient)

\(0.030\pm 0.020\)

Captures strong edge/stripe patterns (e.g., CPB elytral stripes, ladybird spot boundaries), enabling robust species separation.

2

Compactness

Morphological (shape)

\(0.025\pm 0.018\)

Measures body compactness; distinguishes rounded ladybirds from more elongated beetles/larvae.

3

hog_611

Visual (HOG gradient)

\(0.021\pm 0.022\)

Encodes fine edge orientation changes relevant to insect contours and wing/segment structures.

4

Circularity

Morphological (shape)

\(0.022\pm 0.017\)

Quantifies how circular an insect body is, strongly linked to species-level morphology.

5

hog_1294

Visual (HOG gradient)

\(0.020\pm 0.008\)

Detects high-frequency gradient cues useful under variable lighting and backgrounds.

6

Aspect Ratio

Morphological (shape)

\(0.019\pm 0.014\)

Separates slender insects (aphids/larvae) from broader-bodied adults (CPB, ladybird).

7

ALR (Area–Length Ratio)

Morphological (shape)

\(0.019\pm 0.008\)

Combines size and elongation; useful for differentiating developmental stages and species.

8

Roundness

Morphological (shape)

\(0.018\pm 0.013\)

Measures boundary smoothness and body roundness; key for ladybird vs. beetle contrast.

9

Elongation

Morphological (shape)

\(0.017\pm 0.012\)

Reflects body stretching; improves discrimination where color is similar but shape differs.

10

Std_hist

Statistical (color/brightness)

\(0.016\pm 0.017\)

Captures color/brightness dispersion, aiding separation of red-spotted ladybirds from pale aphid colonies and leaf backgrounds.

11

Texture Entropy

Statistical (texture)

\(0.015\pm 0.011\)

Measures texture randomness; helps detect aphid colony granularity vs. smooth leaf surfaces.