Contextual xG and Machine Learning

If you’re still with me after the title, chances are you’re intrigued about xG in a way we’ve rarely explored before. So, what exactly is xG? Simply put, it’s the probability of a shot resulting in a goal. Is it effective? Well, in most cases, yes. It sheds light on the qualities of certain players; for instance, Son Heung-min consistently surpasses his xG tally in the Premier League, showcasing his exceptional finishing ability.

But amidst the praise, there’s skepticism. Many argue that xG is just overcomplicating football with its expected analogies in every model.

But let’s not brush it off so quickly. xG offers a deeper understanding of the game, transcending traditional metrics like goals and assists. It takes into account various factors such as shot location, angle, and type of play leading to the shot, providing a nuanced assessment of scoring opportunities.

However, we are here to deeply understand and figure out if there are some factors involved in the model behind xG that lower its accuracy. It doesn’t consider defensive pressure, goalkeeper positioning, the quality of the ball, the shape of the body when a shot is taken. It also doesn’t take into consideration the chances that were created but probably didn’t lead to a shot and hence missed into the data of xG, and so on.

There are a lot of xG models that incorporate this information into their calculations, making them better versions with more and more data; they only improve. ( StatsBomb’s xG model is a great example )

There has been argument about xG being a complicated stat that doesnt contribute to anything which is absurd statement in itself if you realise how modern football is operating! Data can never surpass the intelligence of human brain but it can help it do things quicker for sure!