A few weeks ago, we concluded that Chelsea was not an unbeatable team, and Chelsea lost the game against Tottenham Spur when we published the analysis next week. Since then, Chelsea seems to be back on the right track and have won 4 out of the next 5 games and remains unbeatable since then. While scoring 11 goals, Chelsea was only scored twice in the last five games. Chelsea is becoming the nightmare for the rest of the teams again as they did in the beginning of this season. Will Chelsea improve to be a perfect team during this period and dominate the league? In fact, based on our research, the defense of the team is still unstable during this period.

According to our statistics model, with 98.94% p-value, if Chelsea does not make any significant adjustment during this period, Chelsea would not be scored in the last three games. In reality, Chelsea was scored in the games against Manchester City and Aston Villa. Especially in the game against Manchester City, the number shown in our research indicates that the team would not be scored in this game, especially with the fact that our statistical model indicates that the performance of Manchester City would not be any problem for the team.

Someone may point out that being scored might be a lucky case in these games. However, according to our new statistic model of Chelsea, the p-value drops from 98.94% to 97.15%, and adjusted R square drops from 57.91% to 37.44%, which are all significant changes, especially we used the same method to generate these models. The result convinced us that there is a significant change in the team defense.

Let us use the Simple K-means and Expected-Maximization to see how the change impact on the team. The data shows no preference in number of clusters we need, so I decided to usse 5 clusters in Simple K-means and 6 clusters in Expected Maximization. These two methods shows similar cluster assignments

Simple K-Mean Clustering,

Game Swansea vs. Chelsea assigns to cluster 4,

Game Chelsea vs. Manchester City and Game Chelsea vs. Everton assign to cluster 2,

Game Aston Villa vs. Chelsea assigns to cluster 3


Game Swansea vs. Chelsea assigns to cluster 1,

Game Chelsea vs. Manchester City and Game Chelsea vs. Everton assign to cluster 2

Game Aston Villa vs. Chelsea assigns to cluster 3

as we can see, both of these two methods agree on the same game assignments. Let us take a look at the results of all the clusters.

Here is a part of the result based on the Simple K-means Clustering,



Here is a part of the result based on Expected-Maximization,


First of all, let us take a look at the results of Cluster 2 based on both methods. For Simple K-Means Clustering, the Goal Against of Cluster 2 is 1.8; for Expected Maximization, the Goal Against of Cluster 2 is 2.3333. These two numbers indicate that the team would not be able to stop being scored in these two games, and will be likely to be scored twice in each of them.

Then we take a look at Game Swansea vs. Chelsea which is assigned to Cluster 4 based on Simple K-Means Clustering and Cluster 1 based on Expected-Maximization. The Goal Against of Cluster 4 based on Simple K-Means Clustering is 0.6; the Goal Against of Cluster 1 based on Expected-Maximization is 0.25 but with standard deviation 0.433. Both of these numbers show that the team is possible to be scored once in this game.

Lastly, we take a look at Game Aston Villa vs. Chelsea which is assigned to Cluster 3 based on both methods. Both of the two numbers, 0.2727 for Simple K-Means Clustering and 0 for Expected-Maximization, indicates that the team would not be scored in this game.

As we can see, at least three out of four games are expected to be scored more than the reality are, we can see that the defense of Chelsea is not invulnerable in the previous four games, but they are just lucky enough to avoid being scored even more. Compared to their performance before making any change, the team is indeed under-performing.

Let us take a look at what changes Mourinho made to his team.

Based on the data, we found that Mourinho required the team to reduce number of set pieces (drop from 0.667 to 0.25), decrease number of clearance (drop from 27.95 to 22), reduce the wing offense (throw ins drops from 18.19 to 14.75), allow the opponent team to make easy passes (opponent passing accuracy increases from 80% to 84%), and try to seek chance outside the box (opponent number of blocks drops from 15.28 to 11.75). In general, the team plays more passively than before. This movement is reasonable, since Chelsea lost to Spur by a fairly big margin. On the other hand, in the long run, this change would not improve the team performance as we prove above.

Even though the results of these four games are fairly good, there is a great potential danger hidden in the satisfied scores. Chelsea is leading 7 points ahead the 2nd place, but they are far from finishing their work this season. If they continue the adjustments stated above, they will encounter significant defensive problems in the coming weeks.


Leave a Reply