FYP Progress (Week 6)

It is mentioned in last week’s update that Multilayer Perceptron (MLP) is being developed. Hence, the algorithm is completed successfully, and the same dataset from previous statistical analytics is used to perform predictive analytics. Everything goes well apart from the fact that the accuracy is too high.

As seen in above confusion matrix, only one match is predicted wrongly for all three possible results, namely home win, draw, and away win. This is caused by the attributes of the datasets, since they contain features that should not be known until the match already ended.

By the time we figure out these numbers (number of goals scored, number of shots, etc.), the match already ended, hence, we should know the result already. Interestingly, one of the literature review for this study uses almost identical features to train their model.

To resolve this issue, the next step to be taken is deriving new features from the existing dataset. The features that will be derived from the dataset will increase or decrease gradually based on teams’ performance. In other words, the values for each features will start with 0.0 and will change as the season progresses.