EPL Analysis and Gameweek 23 Prediction

A data driven attempt in predicting English Premier League results using xG Statistics

This is an article on my EPL Prediction series. You can check out the prediction for Game Week 22 and how it fared against the actual performance here.

Expected Goals or xG is the parameter used for prediction. If you are interested in understanding the algorithm for prediction, I recommend that you check out this article where it is explained in detail.

Analysis up to Game-week 22

After game week 22, Manchester City are on top of the table with 47 points. Pep Guardiola’s side is 3 points clear of the Red Devils with one game in hand.

In the last Game Week, the defending Champions, Liverpool lost to Brighton at home, second consecutive loss at Anfield. Liverpool’s loss was capitalized by Leicester moving into the third position with an away win at the Craven Cottage. The Hammers are at the fifth position beating the Villains at their home. Followed by Chelsea and Everton, who secured 3 points from their away games.

Newly rejuvenated Chelsea taking on their city rivals, Tottenham was the most anticipated match of the weekend. In the match which was hailed as the battle between exponents of two different styles of coaching, it was Tuchel who emerged victorious. Mourinho must be missing his star forward Harry Kane after losing three games in a row. In the bottom of the table clash, the Blades secured a 2–1 win over the Sparrows securing their 3rd win in the last 5 outings.

After 22 gameweeks, the defending champions, despite their recent slump in performance, leads in xG Scored per game closely followed by the current table toppers Man City. Man United who have scored the maximum goals in this season after thrashing Soton 9–0, is at the third place for xG scored. Aston Villa, Leeds and Chelsea have also created more than 1.5xG per game.

Crystal Palace, Burnley, and West Brom, struggling with creativity are at the bottom of the leage in terms of xG scored per game. Southampton following injury woes have been struggling to find the net lately, their xG per game has dropped below 1.

Manchester City seems to the the best defensive unit allowing opponents to create just around 0.75 xG per game. Man City have secured 6 clean sheets in a row in the League.
Most teams concede between 1 to 1.5 xG per game. West Brom and Leeds have conceded more than 1.5xG per game.

Based on xG Scored and xG Conceded, teams can be grouped into 4 quadrants as shown in the above graph.
The horizontal dotted line shows the average xG scored per game. Teams above the horizontal dotted line are strong attacking sides and the teams below, weak in attack.
The vertical dotted line shows the average xG conceded per game. Teams to the left have a strong defense and the teams to the right have week defense.

Man City is head and shoulders above the other teams with respect to delta xG, with a differece of more than 1 between xG scored and xG conceded. Liverpool and Chelsea are the the only other teams with delta xG above 0.5. West Brom on the other hand is struggling at both ends of the pitch with the weakest attack and porous defense.

Leeds United ranks among the top in xG created per match. However, the team has negative delta xG as opponents find it easy to penetrate the Leeds defense, something Bielsa has to immediately look into. Wolves on the other hand is a good team defensively, but they lack the strike force with Raul Jiminez out due to injury.

Game Week 23 Predictions

Before proceeding to the predictions, let me clarify that this is a very simple algorithm just based on past xG, so only baseline performance can be expected. The algorithm also fails to predict high scoring games. The model also does not take into account the team selection, absence of players due to injuries/suspension, formation, tactical changes etc.

However, the model has been performing pretty well in predicting the momentum of the matches. You can check out how the actual performance fared against the predictions made for the previous game week below.

Predictions for Game Week 23 are provided in the table below.
The absolute value of GD shows the competitiveness of the match. The higher the value, more one sided the match is expected to be and higher the accuracy of prediction.
The lower the value of GD, the more the match could be anybody’s game. Positive value of GD means Home win and Negative value means Away team win.

Spurs vs Westbrom seems to be the most one sided match of Gameweek 23. Spurs are expected to emerge from the match with all the three points. Chelsea and Brighton are also expected to comfortably win their away ties against Burnley and Sheffield United respectively.

Leicester and Hammers have a fairly good chance of winning their away matches at the Wolves and Fulham respectively. The Red Devils are also expected to win their match at the Theatre of Dreams against the Toffees.

Newcastle vs Southampton is the most closely contested match of the weekend, where anyone could be the winner. The match is also expected to be a low scorer. Leeds United and Aston Villa can also expect to put up a decent show against their opponents at home.

The most anticipated match of the week is the Super-Sunday clash between the defending champions, Liverpool and the current table toppers, Manchester City. We can expect highly competitive match with the model slightly in favor of Pep’s side. On Sunday we can see whether the addition of new centre backs can turn the tide in Liverpool’s favor.

Update: Predictions Vs Actual Performance

The algorithm was accurate in prediction of 7/10 fixtures of GW23. However, as mentioned earlier, the algorithm fails to predict the magnitude of delta xG.
The fixtures where our predictions went wrong were for 3 home games of Burnley, Fulham and Wolves where the hosts pulled out a performance which was better than the visitors who were expected to perform better. Home advantage factor has to be explored.

Photo by Mario Klassen on Unsplash

Data Analyst | Hacker | Financial Analyst | Freelancer | IIM MBA | Opensource | Democratize Knowledge | https://www.youtube.com/channel/UCLpBd4gzfIBXm2BPpdHOWdQ