Multi-categorical levels time series forecasting with Azure ML Neural Network Regression

Compute time series forecast for number of people entering buildings daily
This experiment train and test a Neural Network Regression model, to predict how many people enter different buildings daily. The model uses Calt2 dataset from UCI Machine Learning repository. The model assumes weekly seasonality. Train model use the data before latest 7days. Test model use latest 7days data. Using t test to determine if two sets of data are significantly different from each other. This is a null hypothesis test. The null hypothesis assumes no relationship between variables in the population from which the sample is selected. In this experiment the result shows the possibility of null hypothesis is 0.006642. Reject null hypothesis is true. So this two sets of data (predict set & test set) are not significantly different from each other. For time series forecasting problem, usually we have to train model for each categorical level because the time series algorism's input only accept time series value column. If you have hundreds or thousands categorical levels you have to train and save hundreds or thousands models. The advantage of this experiment is train one model for multi-categorical levels.