\cite{jolliffe07} Suppose that a set of probability forecasts is made of $n$ binary events of interest such as ``precipitation tomorrow'' or ``damaging frost next month.'' Denote the $n$ forecasts by $\{p_1, p_2,..., p_n\}$, where each $p_i$ is a probability between $0$ and $1$. The corresponding observations $\{x_1, x_2,..., x_n\}$ are coded as $1$ if the event occurs and $0$ if it does not. To assess the quality of the forecasts, various measures or scores can be constructed that quantify the difference between the set of forecasts and the corresponding set of observations. …show more content…
The BS for a sample of $n$ binary forecasts, where the event either occurs or does not occur, is given by
\begin{equation}
\ BS=\frac{1}{n}{\sum_{t=1}^{n}}(p_{i}-x_{i})^2
\end{equation} \vspace{12pt}
\cite{brier50} The Brier score was first introduced by Glenn W. Brier in 1950 to measure the accuracy of weather predictions. It can be applied to all forms of probability forecasting not just weather. The Brier score is the difference between the sum of the squares between forecasts and observations. To demonstrate how to Brier score works, the following example explains it: \vspace{12pt}
Consider a football match, you forecast Team A to score at least once in the match at $80\%$ and they do go on to score in that match. With there only being one forecast, $n=1$ therefore the Brier score is then simply as follows
\begin{equation}
\ BS=(0.8-1)^2=0.04