A central problem in data science is predicting the future of time series data. In recent years many sophisticated methods have been applied to forecast time series data. Neural networks are a subset of machine learning, in which the computer learns to perform some task by analyzing training data sets. In this project, neural networks are trained with the Back-Propagation algorithm for predicting sales and further, it is used for classifying images containing different face portraits. The different optimization methods and activation functions that are used along with Back-Propagation algorithm to train the neural networks have also been discussed.
Keywords: Neural Networks, Back-Propagation algorithm, Time-series forecasting, Activation function, Stochastic Gradient Descent.