# Face features trials ## All face features ![](https://i.imgur.com/5RphP9a.png) | Learn curve | Validation curve | | -------- | -------- | | ![](https://i.imgur.com/6dk9Iei.png) | ![](https://i.imgur.com/U4ZBl7c.png) | ![](https://i.imgur.com/3vo11Mc.png) ## Only the mouth features ### Why? Correlations of mouth indices | Correlation x | Correlation y | Correlation z | | -------- | -------- | -------- | | ![](https://i.imgur.com/mAsfKaS.png) | ![](https://i.imgur.com/S9358pC.png) | ![](https://i.imgur.com/MDE9qZ8.png) | Correlation of eyebrow indices | Correlation x | Correlation y | Correlation z | | -------- | -------- | -------- | | ![](https://i.imgur.com/0pl4LB0.png) | ![](https://i.imgur.com/l3k0tqQ.png) | ![](https://i.imgur.com/LklmZJU.png) | Eyebrows just the same and highly correlated over two parts of the indices (left and right eyebrow probably) ### Execution ![](https://i.imgur.com/5I238le.png) | Learn curve | Validation curve | | -------- | -------- | | ![](https://i.imgur.com/sKllYFr.png) | ![](https://i.imgur.com/QzHefqU.png) | ![](https://i.imgur.com/N5b2eB0.png) We see better confusion matrix for AUTO-RIJDEN, HAAS-oor and SCHILDPAD and this with only 40 features instead of 100 ## All the face features but with w, h and area ![](https://i.imgur.com/UIcQuHz.png) | Learn curve | Validation curve | | -------- | -------- | | ![](https://i.imgur.com/YICFDsT.png) | ![](https://i.imgur.com/Ttu6pfH.png) | ![](https://i.imgur.com/5HaiLTO.png) Not better, took only 80 features instead of 100 and the newly added features were not taken ## Only the mouth indices but with w, h and area Same as only the mouth indices, the newly added features were not selected by selectKBest ## Only the w, h and area features ![](https://i.imgur.com/ELdafpI.png) | Learn curve | Validation curve | | -------- | -------- | | ![](https://i.imgur.com/pNQrFJw.png) | ![](https://i.imgur.com/SBxty7U.png) | ![](https://i.imgur.com/Z1i8dCT.png) Performs "well" with GOED, HAAS and NAAR