Basketball Outcome Prediction And Analyze Based On Conditional Random Fileds

Nowadays, basketball betting market is booming, which also makes the prediction of basketball outcome becoming a challenge. Presently some use the traditional statistical methods to predict the outcome of basketball, such as Naive Bayes Classifier, etc, and treat the basketball outcome problem as a basic binary classification problem. And due to the basic data set, feature selection and the methods, the traditional methods have low efficiency, and accuracy is poor.At present, there is no such unified and quantitative way to describe the strength of the team in traditional methods, in this paper, based on directed graph and PageRank algorithm, we propose TeamRank algorithm to describe the strength of a team. Tested on data from 1990 to 2015 on 30 teams, the result is that TeamRank algorithm can better describe the strength of teams.Conditional Random Fileds is a statistical method based on time series, it can take the basic features of the target into consideration and also describe the context between targets by defining the feature functions. thus it is suitable for the time based basketball data. In this paper, we treat the basketball outcome problem as a sequence label problem by build sequences on a team through the data from 1990 to 2015, and use our carefully selected features along with TeamRank values of the current team as the feature input, we build a model named NBAPrediction based on Conditional Random Fields, and use it on every single team, the result is pretty, can achieve most highest accuracy of 0.7262.