This is a time series forecasting project using Neural Networks.
In this project, I try to make a forecast of criminal activities within the City of london.
Two Neural Networks algorithms will be utilized i.e N-BEATS and DeepAR
The DeepAR model is a cutting-edge time-series forecasting method that creates a probability distribution across the target variable at each time step, enabling it to be used for probabilistic forecasting. The model is trained to maximize the log-likelihood of the observed data given the input time series and categorical variables. This is accomplished through the training process, which aims to make the negative log-likelihood loss function as little as possible. The model uses various gradient descent techniques, such as stochastic gradient descent (SGD) or the Adam optimizer, to optimize the parameters of the DeepAR model. The predicted parameters computed in the model training process are used to obtain the mean and variance of the Gaussian distribution, which can be used for probabilistic forecasting.
NBeats is a sophisticated deep-learning model designed specifically for time-series forecasting tasks. It is based on a layered architecture of fully connected neural networks that can handle long-term dependencies and capture nonlinear patterns in the time series. The NBeats model comprises two primary components: the backcast network and the forecasting network. The backcast network takes the previous data and utilizes those values to produce a set of basis functions. The forecast network then makes use of these basis functions to make predictions about the values of the time series in the future. The model is trained using a loss function that minimizes the difference between the predicted and actual values of the time series. The NBeats model is gaining popularity due to its capacity to capture complicated temporal patterns while simultaneously offering interpretable insights into the process of forecasting.