Using a machine learning approach means letting the computer discover patterns in past weather data. In our approach, we used neural networks, a type of machine learning, to learn these associations.
So what exactly was the data we learned associations from? We pulled hourly data from the weather station database, and by appending hours together, we could learn the associaitions between one hours weather conditions (temperature, humidity, wind spd, etc.) and those of the next hour.
Using WEKA, a machine learning toolkit, we were able to create a neural network to predict each of the next hours weather conditions. Using these predictions, and the previous weather conditions, we were able to extrapolate to get the weather conditions for multiple hours in the future.
Our Neural Network for Predicting Temperature