That's Hot: Predicting Daily Temperature for Dierent Locations

. The problem of effectively modeling weather has been a focus of numeric simulations since the early 1950s. We address the specific task of modeling daily temperature using only temperature data from preceding days and equivalent days in previous years. Weather data compiled by the National Oceanic and Atmospheric Administration (NOAA) for five chosen cities over a period of 31 years was used as training data. An autoregressive analysis generated a temperature hypothesis function with mean absolute error of 2.58◦F against the true temperature values. For validation of the applicability of a temperature-based data assumption, we used the k-nearest neighbors algorithm (with previous days’ temperatures used as coordinates), obtaining a mean absolute error of 2.70◦F . The similar predictions of the two models suggests stronger confidence in the effectiveness of temperature modeling through a singular dependence on preceding temperature data. 2 Introduction