Forecasting Temperature Using AI Linear Models

By Vikram Kavalipati

Abstract

Currently, weather forecasting is done primarily using traditional computer models. Artificial intelligence (AI) would allow predictions to be made much more easily with far less effort. However, AI weather models are still less accurate than traditional computer models. The goal of this project is to explore the possibility of using artificial intelligence in forecasting and to gain an understanding of its potential and limitations while creating an accurate weather forecast. The models use a 70 MB dataset that is in JSON format. The dataset includes a variety of factors including min and max temperature, wind speed and direction, air pressure, and precipitation for more than 50 years. This paper uses a wide variety of regression models to forecast the next day’s temperature for Roseville, CA, including Linear Regression, Ridge Regression, Lasso, etc. Linear regression was consistently the best-performing model, with a mean absolute error of 2.29 degrees, suggesting that even simple linear models can produce weather forecasts that are generally close to the correct value.