Data Preparation (11) - Clean (Feature Selection)

FSelector (Romanski, 2013) 패키지는 주어진 데이터세트에서 속성을 선택할 수 있는 기능을 제공한다. 

관련성이 없거나 불필요한 정보를 확정하고 제거하는 기능을 한다. 

> library(FSelector)

> form <- formula(paste(target, "~.")) // paste() concatenate vectors after converting to character

> cfs(form, ds[vars]) // cfs : algorithm finds attribute subset using correlation and entropy measures for continous and discrete data

[1] "min_temp"        "sunshine"        "wind_gust_speed" "humidity_3pm"   

[5] "pressure_3pm"    "cloud_3pm"      

> information.gain(form, ds[vars]) // information.gain = H(Class) + H(Attribute) − H(Class, Attribute)

                attr_importance

min_temp           3.539250e-02

max_temp           0.000000e+00

rainfall           0.000000e+00

evaporation        0.000000e+00

sunshine           6.523179e-02

wind_gust_dir      4.073802e-02

wind_gust_speed    3.931861e-02

wind_dir_9am       3.537000e-02

wind_dir_3pm       1.759904e-02

wind_speed_9am     9.813415e-05

wind_speed_3pm     0.000000e+00

humidity_9am       2.858310e-02

humidity_3pm       6.189702e-02

pressure_9am       5.317622e-02

pressure_3pm       6.878745e-02

cloud_9am          3.314110e-02

cloud_3pm          6.893149e-02

temp_9am           0.000000e+00

temp_3pm           0.000000e+00

rain_today         1.261390e-02

>