Big Internet giants have access to the personal data of more and more people. This already created a mounting attack from goverments in Europe and USA. Even the proverb says "Birds of feather flock together" so it is relatively easy to guess the properties of an individual based on its friends even if one wants to hide it. It is important to know to what extent internet companies can guess our features. The task of this thesis is to go beyond the easy predicting like label propagation and use methods from network science and artificial intelligence to do better. The major aim would be to implement a convolution neural network type approach on networks. The major challenge in the social egocentric network of individuals is the varying number of friends which have different relation to the ego. So first one must find the relevant partners in our egocentric network and apply the detection method on them. Then use the neural network approach to get a good trait prediction.