Fig. 3
From: Exploration of transfer learning techniques for the prediction of PM10
Random forest feature importance. Each subfigure visually represents the significance of individual features, showing the percentage by which each feature explains the PM10 concentration. Features include temperature, day of the year, relative humidity (rh), station ID, weekend, holiday, day before and after holiday, wind speed (windsp), wind direction class (windDirClass), precipitation (precip), pressure, and lagged PM10 value. (a) Depicts the city-level model of Graz without the PM10-lag feature, while (b) includes this feature. (c,d) focus on a single station in Zagreb (station-level) using the same approaches. These figures highlight the significant influence of the lagged PM10 feature, indicating its explanatory power on the PM10 concentration of the subsequent day.
