In this project I tried to create some tools in order to help a general manager to build a basketball team with the features that he needs. I split the project in two parts, at the first part I created a recommendation model which present players to the manager based on the features that he defined and at the second part, I tried to build a prediction model aiming to predict the position of the player based on specific features.
I created the recommendation model using the K Nearest Neighbour technique. I started with the hypothesis that as a general manager of the Milwaukee Bucks I was looking for a good fit for Antentokoumbo to create a good team aiming to win the championship. Giannis has a very strong skill package but needs more players to support him.
Giannis needs a player that can create for him and also take the advantage of his creative ability. So I'm looking for a young ambitious players with offensive skills.
I started my searching with these features: age=27, 3P%=0.40, 3PA=6, 2P%=0.45, AST=6, TOV=3 and I specify the number of neighbors as 3. The model recommend three players and one of them is Cris Midletton currently Giannis teammate. Despite that Middleton is a nice shooting guard, I tried to make my search more specific in order to pick a point guard.
From the column position I converted the categorical variables into dummy variables and also in my second target group I added the salary feature. The recommendation results were interesting, the first player was Malcom Brogdon former Giannis teammate that Milwaukee chose to trade, the second one was Terry Rozier a player with potentials who had a great playoff series last year against Bucks. The third prediction wasn’t a point guard however the recommandations were acceptable.
The second part of this project it will be a classification moldel aiming to predict the player position..