Precipitation Prediction Using Multiple AI Approaches

Timothy Haskins, Emma Doyle, Reid Hoffmeier, and Ning Yu, Department of Computer Science, SUNY Brockport, 350 New Campus Drive, Brockport NY 14420

Humanity has been recording weather data and attempting to predict it for thousands of years. As a result, there is a plethora of weather data and even more ways to express that data. This project aims to take advantage of the availability of that data and modern AI to determine the best combination of data expression and algorithm for predicting if the accumulated precipitation for Rochester, NY will be greater than zero inches for a given day. Based on the research of similar projects, the accuracy goal for this project is above 70%. Ten years of daily weather data were calculated from weather stations in and around Rochester, NY. The data describes the distribution of different measurements such as wind speed, temperature, dew point, relative humidity, and wind direction for a given day. Multiple versions of the data are derived from the original dataset. There are purely categorical versions, purely quantitative versions, and hybrids of the two pure sets. There are also versions of the data that have been normalized with different normalization methods and some that are not normalized at all. Each of these different data expressions are run through different algorithms to determine which works best for each expression. The algorithms used in this project are: K-Nearest Neighbor, Deep Neural Net, Wide Neural Net, Deep and Wide Neural Net, SVM, LSTM, and Transformers. For all of the different versions of data, each is split into a training set and a testing set. The training set is 80% and the testing set is 20% of the data for each set. After an algorithm has been trained or fit to the training sample of the set, it is then set to predict the precipitation category for the test set and compared to the known precipitation to calculate accuracy.

Additional Abstract Information

Presenter: Timothy Haskins

Institution: State University of New York- Brockport

Type: Oral

Subject: Computer Science

Status: Approved

Time and Location

Session: Oral 9
Date/Time: Wed 12:00pm-1:00pm
Session Number: 913
List other presenters in this same room and session