SHAP Analysis

See how the model makes price predictions and which features push the value up or down.

Model

Summary importance (mean |SHAP|) — LightGBM · TF-IDF+SVD — Summary importance (mean |SHAP|)
Each feature’s average contribution to price — larger = stronger driver. · LightGBM · TF-IDF+SVD

Beeswarm (value spread) — LightGBM · TF-IDF+SVD — Beeswarm (value spread)
Each dot is a listing; colour = feature value (red high), horizontal = price impact (right = pushes up). · LightGBM · TF-IDF+SVD

SHAP computed on all data (no sampling). Group SHAP = signed sum of member SHAPs, then mean-absolute. Free text (beyond model+series) contributes ~nothing — “text adds ~0 to price”.

How to read this chart

Horizontal axis: impact magnitude. Right increases price (+), left decreases (–).
Red dots: a high value for that feature (e.g. high horsepower).
Blue dots: a low value for that feature (e.g. low mileage).

Example scenarios

Mileage (km)

Red dots (high km) cluster on the left — high mileage lowers the price.

Engine power (HP)

Red dots (high HP) sit on the right — a powerful engine raises the price.

Brand & model

Categorical, so dots are gray — read position, not color: right = premium, left = value-lowering.