Se retoma el modelo de outliers
Primero eliminamos los atributos que generamos en la deteccion de otuliers y utilizamos el modulo Correlation Matrix
Podemos observar que los atributos con mayor correlación son:
| Atributo 1 | Atributo 2 | Correlación |
|---|---|---|
| radius_mean | perimeter_mean | 0.997561623039936 |
| radius_worst | perimeter_worst | 0.992657065474132 |
| radius_mean | area_mean | 0.9919003449129997 |
| perimeter_mean | area_mean | 0.9902080430937652 |
| radius_worst | area_worst | 0.98999500432762 |
| perimeter_worst | area_worst | 0.9824612396754517 |
| radius_mean | radius_worst | 0.9801099075417596 |
| perimeter_mean | radius_worst | 0.9800989345342787 |
| perimeter_mean | perimeter_worst | 0.9798591412747525 |
| area_mean | radius_worst | 0.9793034894447585 |
| area_mean | area_worst | 0.9769836451092457 |
| radius_mean | perimeter_worst | 0.9737641514538055 |
| area_mean | perimeter_worst | 0.9731997180741648 |
| perimeter_mean | area_worst | 0.9618158006283064 |
| radius_mean | area_worst | 0.9613407449790952 |
| radius_se | area_se | 0.9588042517676202 |
| radius_se | perimeter_se | 0.9581078557703023 |
| perimeter_se | area_se | 0.9286230267439933 |
| texture_mean | texture_worst | 0.9185988050092276 |
| concave points_mean | concave points_worst | 0.9185319555861766 |
| concavity_mean | concave points_mean | 0.9109626681857491 |
| concavity_mean | concavity_worst | 0.9040835104706433 |
| compactness_worst | concavity_worst | 0.8892615552044842 |
| compactness_mean | concavity_mean | 0.8822007444342892 |
| area_se | area_worst | 0.8767762149945185 |
| compactness_mean | compactness_worst | 0.8708906979751478 |
| concavity_mean | concave points_worst | 0.8581661018635018 |
| concave points_mean | perimeter_worst | 0.8526946702369379 |
| area_se | perimeter_worst | 0.8515309069167041 |
| concavity_worst | concave points_worst | 0.8500507009197358 |
Podemos observar que las correlaciones mas altas son entre areas, perimietros y radios.
Podemos ver claramente que estos atributos estan correlacionados, por lo cual eliminaremos los atributos de la segunda columna con una correlacion superior a 0.85. Para esto usamos el modulo Remove correlated attributes.
