In the previous step, we converted the image into a feature vector and now we will see how the classification algorithm takes this feature vector as input and identifies the object accurately. To make sure that our algorithm works with high accuracy, we need to train it with large volumes of data that contains the object to be identified. If the object to be identified is a cat, for example, then we will feed the algorithm training data of cat and non-cat images.

The principal of machine learning algorithm is to treat the feature vector as points in higher dimensional space. Then, it finds out the planes and surfaces that separate higher dimensional space, which enables it to separate the object to be identified from the rest of the image.

To build a predictive model around this, we need a neural network system. The neural network system consists of large number of a interconnected nodes, and each node is a combination of hardware and software. The neural network uses one of the many classification algorithms available, such as bag-of-words, Support Vector Machine (SVM), face landmark estimation, K-Nearest Neighbors (KNN), and logistic regression.