In this paper we formulate the problem of predicting the outcome (winner) of an ongoing National Basketball Association (NBA) match as a supervised machine learning problem. In this approach, as the match progresses, the outcome prediction dynamically adapts to the current match situation which does not happen in static prediction problems. Dynamic outcome prediction of the match seems to have received less attention in the literature than static predictions. We compute the features used in the classification problem based on the most up-to-date match situation till that point in time. A total of 32 features are used. We present results for all the regular season matches of the 2022–23 NBA season. Dynamic prediction accuracy varies from about 62% at the beginning of the match to about 78% at the beginning of the final quarter of the match. We also evaluate feature importance in this classification problem.