Being involved with data science and simulations, some years ago the idea evolved to write a programme that simulates and predicts badminton matches.The input to this simulaiion would be the information who played against who and what the result was. The output would be threefold: First a prediction who will win a certain rally in a past or future match. Second a prediction who will win a match between two certain players or pairs. Finally a ranking, giving an ordered list of the players such thateach player has a probability higher than 50% of winning against each player ranked below him or her, and a probability of less than 50% of winning against each player or pair ranked above. Thus for the ranking, each player is assigned one number, the strength, used to order the ranking.


Each discipline is handled independently. Thus players will have three strength values, one for singles, one for level doubles and one for mixed doubles.

Used matches

One question is how far back in time are past matches included to simulate the strengths or a match on a particular date? Choosing a time frame that is too short could lead to too few matches used for assessing the players. Choosing a time frame that is too long could mean including matches that don’t reflect the current strengths of the players, because they have evolved since these matches.

While implementing the simulation it emerged that including the last 25 months is a very good choice. Two years as a time frame to establish a good estimate. The additional month comes in handy in order to include tha last two editions of yearly tournaments.

When a strength value is given, it is usually accompanied with the number of matches in the previous 25 months that were included in hte analysis.


The normalization is chosen such taht the average top 10 players or pairs have an average strength of 21 and the other players have a strength that is equal to the average number of points they would win in one game against an average top 10 player or pair.