2. What is the data that we need?

In the first part, I used underlying statistics of the previous 4 games to predict a player's performance in the next 4 games. But every Premier League player is different. If Aubameyang and Ben Mee have the same underlying statistics for 4 games, it does not mean that they are equally likely to score in the next 4. So it's very important to include some long term features for each player too.

Another aspect of how well a player is likely to score or assist is the defensive strength of the opposing team. Including this will not only help us predict whether an attacking player is likely to score but also whether a defender is likely to keep a clean sheet. 

I'm in two minds about including the attacking strength of the player's own team in the feature set. If the individual player's attacking stats are a part of the feature set, I feel there's no need to include the team's overall statistics. That correlation can be measured and revisited at a later time. A team's overall attacking statistics can however be used to predict whether an opponent will keep a clean sheet or not so it can be used there. It's important not to add too many features from the off so I'll start simple and add more features as needed.

Thus for each attacking player, the new feature set is going to look like this: 

Attacking Player Features: 

• Relevant attacking statistics in last 4 games (xG, xA, goals, assists)
• Relevant attacking statistics in long term - 19 games? (xG, xA, goals, assists)
• Opponents defensive statistics in last 4 games (xG, goals)
• Opponents defensive statistics in long term (xG, goals)

Attacking Player Label 1: Likelihood of scoring/assisting in next 4 games

Attacking Player Label 2: Number of attacking returns in next 4 games

Defensive Player Features: 

• Own teams defensive statistics in last 4 games (xGC, goals conceded)
• Own teams defensive statistics long term (xGC, goals conceded)
• Opponents attacking statistics in last 4 games (xG, goals)
• Opponents attacking statistics in long term (xG, goals)

Defensive Player Label 1: Likelihood of getting a clean sheet in next 4 games

Defensive Player Label 2: Number of clean sheet returns in next 4 games

Every defender will also feature in the list of attacking players but not the other way around as only defenders are eligible for clean sheets. 

So we should end up with a list of players and predicted number of returns in the next 4 games. 

Next steps

• Scraping data from Fantasy Football Scout members area using Bs4 in Python
• Home/away statistics