Forest Estimation for Forecasting Football Results

Accurate predictions for a machine learning algorithm are a big decisive factor whether or not the technique sees any empirical use. Moreover, over the last decade an increased interest in football (soccer) has also cultivated to a similar increase in interest in the forecasting of the results. One of the prediction techniques that has gained ground is the use of random forests and variants based on it. In this research the variant introduced by Goller et al. (2018), the ordered random forest algorithm, is used with a database of variables based on past seasons, to simulate the outcomes of every match in the latest German football seasons. The given points are accumulated to make accurate predictions on the final ranking. Additionally, we introduce the neural ordered random forest technique as hybrid on the known algorithm ordered random forest and a neural network, by mapping every tree within the forest to a neural network structure. Although, it's performance is outclassed in some areas compared to the original technique, multiple adaptions are suggested for future improvement.