Fantasy Soccer Recommendation System

Overview

A nerdy, competitive soccer fan has recently been introduced to the world of Fantasy Football. As a new manager of a virtual soccer team, he faces an immediate goal of building a team that competes and wins in his ‘league’. A league here is his group of friends or peers who are competing with him. Each week, his basic objective is to ensure his team gets the most Fantasy points by making good player selection decisions. There are several constraints to making a good team which include the budget that is allocated to each team in the league, and the limited pool of available players for selection.

An organizing system, therefore, is required to support the goal of building a winning team by choosing a mix of players that will perform well in their respective matches and potentially generate maximum Fantasy points. While there are several different ways to achieve this, I am going to tackle this problem by organizing the information about the players in such a way that makes the selection process cognitively less demanding and more intuitive for the new Fantasy Soccer manager.

What resources are being organized?

A Fantasy soccer team as an entity can be represented at two different levels of abstraction. The first and more abstract level is the ‘squad’ of players available for selection. The squad typically consists of 17 players. The second layer of abstraction is the ‘playing eleven’ for a particular week. The squad is selected from a common ‘universe’ of players that is available to everyone in the league. Both these levels have have specific rules and place different constraints that determine the inclusion of a player. For example you cannot choose more than 3 players from the same club or team to be part of your squad.

We are interested in organizing information about the universe of players so that a manager can easily and intuitively select a squad and a playing eleven for an upcoming match. In any case the resources being organized are two kinds of digital resources, i.e.,  soccer players(virtual versions) and information about those players.

This set of information (or description resources) about players basically consist of performance statistics that can be organized to analyse trends and generate predictive, data-driven models to assist the selection process.

Why are the resources organized?

Both kinds of resources are organized to assist in the singular goal of decision support. In particular the decisions that a virtual manager makes each week will include ‘keeping’, ‘benching’, and ‘trading’ players. So the primary resources must be organized into these three categories. The only way the primary resources can be organized is by first selecting and organizing their description resources.

For anyone new to the world of Fantasy sports, it can be an overwhelming initial experience and furthermore, it can be years before people learn to recognise and analyse the data available to them. This organizing system is meant to somewhat smoothen the learning curve.

How much are the resources organized?

The primary resources in this organizing system are players. Ontologies in soccer management are typically well defined. But to be more explicit, at the lowest level of granularity, players are classified by teams they play for, and then by the positions they play in.

For the purpose of Fantasy football,  any type of organizing system will include these two basic classifications. One way of further increasing granularity is further classification. So, if a player X belongs to team Y and plays as a “Midfielder”, the position could be further classified into “Central” and “Wing”, and then “Central Midfielders” could further be classified into “Attacking” or “Defensive”. This would present a hierarchical organization scheme.

For our system, several deeper levels of granularity would be required for each description resource. “Central Attacking Midfielders” would have their own description resources like “number of passes”, “number of passes”, “number of goals”, “number of assists”, “work rate(distance run)”, etc. along with their averages. Depending on some logic applied to these description resources, a “central attacking midfielder” could then theoretically be classified into the “Keep”, “Bench”, or “Trade” categories.
Alternatively, the same description resources could classify the midfielder into “playmaker”, “scorer”, and other intermediate categories before finally classifying them into “Keep”, “Bench”, or “Trade”.

When are the resources organized?

Organizing begins when a Fantasy manager joins a league, so that he is able to pick a squad initially. The activity of organizing then continues throughout the season, and happens after every match. A player might not have been recommended initially, but may have improved many-fold over time to warrant selection, or a player might perform badly over time, or a player just might not be the best option in the position for the upcoming match. In any given case, the players must be organized continuously to accurately recommend players.

Who does the organizing?

Since the goal of the system is to assist decision making, it is only natural that the system does the organizing. However, the system does not enforce its recommended scheme. The Fantasy manager has the flexibility to select the players and ignore the system – in which case, the selection principles might be arbitrary.

Other considerations

An important consideration in such an organizing system is maintenance. What happens when there another parameter becomes available which can be measured to increase the accuracy of the predictions?

Another consideration in organizing players in Fantasy soccer is the selection of ‘formations’. A formation is very crucial in influencing the number of points a Fantasy team gets each week. As of now this system only tells the manager which players are likely to generate the maximum points for their positions. The user should be able to select 11 players that will potentially generate the maximum points and arrange them into an allowed formation.