This project is property the Serendeepia Research company , which wishes make avaible the results obtained by the students of the artificial intelligence magister of the University Internacional de la Rioja to all community . The main target this virtual carer is detect the emotions in the facials expression to star a therapy agree the emotions that show the teeanagers that suffer this disorder.


Introduction to Body dysmorphic disorder


This disorder is defined by (Sandoval O, García-Huidobro R., & Pérez-Cotapos S., 2009) as "the excessive worry  and  disproportionate for a minimum defect or imaginary  in the physics appearance. " (p.01) Likewise, the authors indicate in their post that the patient must have a boost the social relationship,the adjustment of his reality and the acceptance of himself.

Agree with previous, 
it was proposed work with artificial vision to monitor the 
patient detecting the emotions with  he purpose of support his care in 
hospital or in the his house. It is a good method why the patient show emotions
face a mirror presenting negative emotions of himself with which it can alert to
family or send message his mobile boost the social relations or go up his 
self-esteem.


Context

For work with artificial intelligence algorithms, it is have two options , a of this  is machine learning and the other is the deep learning who working with supervised learning and unsupervised learning. in this case it is working with tensorflow with datasets who answer the paradigm of the unsupervised deep learning.


Choose and download dataset

To start, the procedure that it show involves two standpoints where it training with datasets for Facial Expression Recognition and Facial Keypoints Detection, you can dowanload them here:


facial expression recognitionfacial keypoints detection 


This datasets were find in keggle, is the most important  competitions   web page the in data science and the most complete. The search of datasets is one of the main works to train the IA models. Whitout big datasets is very hard make that the machine learn. 

Explore Data


once the dataset is selected and imported into the notebook python , the content must be  explored and understood. In this case the emotion dataset content:





It look that the dataset is completely clean, it have labels that identify which are the data of training, of test and of private test. Likewise it is noted that the dataset have seven labels of the a which the label "Disgust" there are five thousand forty seven records, this is very little compared with the rest labels and maybe the machine learn bad this example. The dataset must change the format at present the information have a dimension of (35887 , 3) for the model it must a dimension of (35887 , 48 ,48) that only have the image with the real size, the column  the emotions is a set aside that the algorithm of tensorflow relate only in the training. 


Changing the dimensions for the model





This step is very important why without the change the model don't understand the data. Now that the information have the correct format the trainig and test dataset are created next model:



The creation of this model was very hard why it started since one neural network artificial simple who did not understant the information and it was spending long time, the next model it was neural network multilayer this it had better result but it was enought, so that it was created a model convolutional:



This convolutional neural network apply filters to images the dataset with the purpose of extract more features, this procedure is the better later the 10 epochs it reach the goal of 68% accuracy 




























*