Analysing Playing Style & Carrying Out Principal Component Analysis To Find Pattern In ISL Teams

Mumbai City - Mumbai had the least number of passes and touches for a side in the 2019-20 ISL season. However, they still did pretty well as they finished 5th and missed out on playoff spots by three points. They relied heavily on crosses and wing play. The Maharashtrian outfit had the least passes/cross rate at 22.34. Their touches/ cross rate stood at 33.49, which implies almost their every 34th touch was a cross. Their forwards were also willing to take shots as they had a pop at goal at the rate of 27.37 passes/shot, which is the least in the ISL.

Teams in first quadrant(ATK,HFC, FC Goa) are relying less on crosses. Clubs in third quadrant rely heavily on crosses. Chennai, Bengaluru and N.E United are all round about the average mark.

FC Goa- Relies on short passing and patient build-up. Had the most number of touches and passes for a side in the ISL season. Their average possession of 58.1% was the highest in the league. Defensively, could have been better as they conceded as many chances as 5th placed Mumbai City FC. Goa conceded 226 chances and Mumbai 225. However, the table toppers clearly dominated the chance creation part as Goa had the highest number of goal scoring opportunities created at 294. They didn't rely on crossing too much as they had the worst passes/cross rate, which stood at 40.42.

Interesting to note that the top 4 sides for 2019-20 season, all finished above average on chances created and xG for parameter.

ATK- A team which stood up on the big occasion and made their chances count. The stats also say the same as they had to make the least number of passes to create a chance. Their passes/chance creation rate stood at 27. Goa who are primarily considered as the most eye-catching and best attacking team in ISL, they took 32.72 passes to create a chance. The three-time winners depended on swift counter-attacking and direct football. Their underlying defensive numbers aren't too impressive though as they conceded 261 chances throughout the season. Only Hyderabad FC conceded more chances than them at 288. ATK might have had the second-best defensive record in the league(in terms of goals conceded), but they did allow a fair amount of chances. However, if one looks at their xG conceded, they had the 4th best number at 26.1(here best-being, a team which has a lesser value), which implies they conceded a healthy amount of chances, but they were of low quality. Habas' side didn't rely on crossing too much, and their style was to counterattack from the middle.

Kerala Blasters-

Kerala stood out for conceding the least number of chances at 167. The next on the list in terms of chance conceded were Bengaluru at 205. Blasters' xG against stood at 23.1 which is again the best in the league. However, they let in 32 goals in the season. Only Hyderabad, Chennai and Jamshedpur conceded more than them. With their underlying defensive numbers being so good, one can classify them as a bit unlucky at the back.

In terms of attack, their expected goals for the season stood at 25. They took the second-highest numbers of passes to create one chance as their passes/chance created value came at34.87. The team also relied on crosses to create opportunities. Their average possession (56.44%) for the season was only behind FC Goa(58.1). The club was guilty of being too passive in possession, and it lacked the cutting edge. When they were not able to find a breakthrough from the middle, they relied on crossing the ball.

Odisha FC- The side had the third-highest average possession for the season. However, like Kerala, they were too passive on the ball and lacked penetration. Only NEUFC created lesser chances than them throughout the season. Odisha managed 206 chances for the last campaign. The side didn't rely on crosses, and their general philosophy was to build up from the back. They had to make the highest number of passes(34.92) to create a chance which shows their lack of ideas in the final third.

Their chances conceded stood at 240. The likes of Hyderabad, ATK, N.E United and Chennai conceded more chances than them. However, Odisha's xG Against was the second-worst in the league. It means they might have conceded lesser chances than few clubs, but they were conceding high-quality chances.

Bengaluru FC- The stat which stands out for BFC is their xG difference. Bengaluru should have scored around 34 goals in the season given the chances they got, but they only managed 24. Their xG difference of -9.4 is worst among all the teams. Defensively, Bengaluru conceded only 205 chances in the season which is the second-best in the league. Bengaluru had a balanced approach in terms of playing style. Their passes/cross is just about average for the season. One can put them in the category of a 'reactive side' as they adopted their game according to the opposition and did reasonably well even though they were let down by their finishing.

Jamshedpur FC- A team which showed promise early on in the season as they lost only one of their first 8 games. JFC were making most passes to take a shot at goal (35.22) along with the most touches/shot. This implies they were indirect in their approach. Their variance of possession over the course of the season was also quite high which indicates that they didn't have a nailed on playing approach. Against Mumbai City in one of the games, JFC hogged the ball and had 61% possession, whereas in the other game against the same outfit, they managed 38% possession. JFC also relied on crosses a fair bit as they had the third least passes/cross rate. Only Kerala, NEU and Odisha had a far worse xG creation rate which showcases their inefficiency in attack. Defensively, they were roundabout average.

Chennai FC- A side which peaked at the right time and almost won the title in the end. Chennai went undefeated in their last eight league games. They had a direct playing style as they took 27.57 passes to create a chance. After FC Goa, they created the most chances in the season. Their touches/shot stood at 41.62, which was the third least in the season. Their crossing numbers were average which shows that they were ready to mix things up by putting in crosses from time to time. Defensively, they could have been better as only league strugglers like Hyderabad FC, North East united and Odisha FC conceded a higher amount of xG.

Hyderabad FC-

Hyderabad conceded the most chances for the season and not surprisingly conceded the most goals (40). They were direct in their playing style though as their passes/shot rate of 27.48 is the second least for the season. First being Mumbai City at (27.37). Their touches/chance creation number(39.94) was the best in the league, which is quite surprising. Hyderabad created more chances than the likes of Kerala Blasters, Jamshedpur and Mumbai City FC.(See above chances created graph)

Only FC Goa took more touches/ cross than Hyderabad FC which indicates that Hyderabad clearly didn't rely on throwing in crosses in the box. One can classify them as a counter-attacking side which relied on direct passing through the middle to move up the pitch. Their underlying attacking number isn't too bad, and If they can sort out their defensive vulnerabilities, Hyderabad could emerge as a far better side.

Principal Component Analysis

To sum things up, i decided to carry out a Principal Component Analysis. I have thrown a lot of factors like passes, crosses, chance conceded, touches, xG created etc to analyse the teams above. PCA can turn all the variables into a 2D graph and can indicate a story in just one picture.

A simple way to interpret the below graph is that club names which appear closer together are similar in style. The red line indicates which stat influenced their position. With scatter plot, you can only plot two variables, but with PCA, you can make sense of more than two variables. I have used 11 variables here.

The angle formed by two variables (red lines) also indicates the relationship between those variables. Example, the angle formed by goals for and touches is small, that means they are positively correlated. Similarly, the angle formed between saves and passes is obtuse, and they are negatively correlated. An angle closer to right angle indicates there is no correlation between those variables.

Findings from the graph

• Goa is undoubtedly ruling the roost as they have impressive attacking numbers (You don't need PCA to know that though). Chennai can be termed as the second-best attacking team as they are closest to Goa.
  

• Odisha FC, Mumbai City and Jamshedpur's numbers are quite similar to each other across most of the variables and that is why they are clustered together.

• Likewise, ATK and Chennai can be said to have a similar style as they are close together.

• Teams which are more closer to the centre have a balanced number across all the 11 variables. That is why FC Goa and N.E United are almost outliers as one has an outstanding attacking number and one has a weak attacking output.

Drawback

• PCA is not a hundred percent perfect, and there is some amount of information lost because of numerous variables. In my graph, it explains(78.46%) of the variance based on chosen variables.

• You might think that, Mumbai City should have been much closer to the crosses variable as they put in the most crosses. However, PCA tries to form the best fit over all the variables.

More explanation(You can skip this if you want to)

Goa is attracted towards shots, chances created, xG created etc variables because it has impressive numbers in all those parameters. Think of it as Goa is being pulled towards the far right of the graph by those 6 variables.

Mumbai will be on the opposite side of Goa because its numbers aren't too great on those 6 parameters. MCFC had the least passes and touches and their attacking numbers aren't really the best.

The only thing pulling Mumbai towards the right side of the graph is crosses as it has a high value over there. However, they are not getting too close on the right because only one variable is pulling them towards the right(i.e the cross) and the other six variables are working against them to send it on the other side. That is why we don't see Mumbai close to crosses as PCA tries to find the best fit across all the variables.

Nevertheless, we still see Mumbai being closer to crosses than Odisha & Jamshedpur.

As mentioned, PCA is not perfect but it does give a roundabout fair idea about a team as it can visualise various factors in an interpretable 2D graph. PCA can also be beneficial in player recruitment, especially if a team is trying to replace a player and wants a like for like replacement.

Note: All data have been taken from fbref via statsbomb, ISL's official website and greyareaanalytics.