Individual_project (PGA cut predictions)

Goal:

• This data science project will attempt to predict whether or not a player will make the cut, based on previous weeks performance, coupled with season averages(drive distance, fairway percentage, putting average).

Description:

• Is it possible to predict 'making the cut' or not at a PGA event?

• Making the cut is important for players and sponsors alike.

  • A typical PGA event is 4 rounds of golf. Thursday - Sunday
  • 'The top 70 players (including ties) make the cut. Any player in 71st place or worse (after Friday's round) is cut.' - pga.com

• From a players perspective, making the cut is important for many reasons.

  • If a player misses the weekend, they do not get paid.
  • Also, prized FEDEX cup points are missed. ('The FedEx Cup is the central professional golf competition that compares player performance across part of the PGA Tour season in a League and Playoffs structure.')
  • For continued poor performance, their tour eligibilty could be in jeopardy.

• From a sponsors perspective, making the cut equates to 'eyes on product'.

  • Players making the weekend are televised more. Sunday draws the largest viewership of the tournament days.
    • 2-4 million viewers for non-majors
    • 5 million for majors
    • 10 million for 'The Masters'

Data Dictionary

Variable	Description
date	Date of the event
course	Name of the golf course
tournament name	Name of the tournament
made_cut	Indicator of whether the player made the cut or not
sg_putt_prev	Strokes gained/lost putting in previous events
sg_arg_prev	Strokes gained/lost around-the-green in previous events
sg_app_prev	Strokes gained/lost approach in previous events
sg_ott_prev	Strokes gained/lost off-the-tee in previous events
sg_t2g_prev	Strokes gained/lost tee-to-green in previous events
sg_total_prev	Total strokes gained/lost in previous events
driving_avg	Distance of the player's drives in yards
fairways_hit	Percentage of fairways hit by the player
putting_avg	Average number of putts per hole

Project Planning:

• Acquire a dataset that peaks a personal interest - Golf
• Prepare/clean the data for exploration
• Split data into train, validate and test
• Explore the data for significant features and answer questions we have about the data set
  • Perform statistical tests that follow assumptive guidelines
    • number of observations > 30
    • test for equal variance and ensure 'equal_var' is properly set based on Levene's test
    • normal distribution (using the central limit theorm for this study)
  • Allow the data to guide us.
• Model on scaled train,validate data sets
  • Run model for accuracy on test dataset to see if there is predictive power
    • Predict whether or not a player will make the cut at a PGA event
• Make recommendations and formulate next steps to continue the project

Initial Questions:

• Do strokes gained putting in the previous week affect whether or not someone will make the cut this week?
• Do strokes gained around green in the previous week affect whether or not someone will make the cut this week?
• Do strokes gained approach in the previous week affect whether or not someone will make the cut this week?
• Do strokes gained off-the-tee in the previous week affect whether or not someone will make the cut this week?
• Do strokes gained tee-to-green in the previous week affect whether or not someone will make the cut this week?
• Do strokes gained total in the previous week affect whether or not someone will make the cut this week?
• Does driving average affect whether or not someone will make the cut?
• Does fairways hit percentage affect whether or not someone will make the cut?
• Does putting average affect whether or not someone will make the cut?

Key findings:

• Players that make the cut gain .25 strokes on the field in 'strokes gained approach'
  • 'Approach measures a players shots after the initial drive, so it highlights which golfers have excelled in those crucial second and third shots.'
  • Second and third shot (accuracy) is a premium stat when trying to make the cut
• Players who make the cut gain .50 strokes on the field in 'strokes gained tee-to-green'
  • 'Tee-to-Green is perhaps the best metric of them all. It’s essentially Total Strokes Gained — which is the sum of off-the-tee, approach, around-the-green'
    • This stat covers complete game. Tee ball accuracy, iron play, and chipping around the green.
    • If players are lagging in this stat, it may uncover weakness in their game and provide insight for focused practice
• Driving average/Putting average
  • Those that make the cut are 2 yards longer on average (296 vs. 294)
  • Those that make the cut average 1.76 putts per hole vs. 1.77 of those that do not
    • Golf is a game of inches and on any given Friday, the total game is required to make the cut
      • And.... maybe a little luck

Steps to reproduce:

• Aquire data from 'https://zenodo.org/record/5235684'
  • save to your working directory and read into a notebook using pd.read_csv(name_of_file)
• Prep the data for exploration:
  • using function 'prep_pga()' in the prepare.py file included in the Github repository
• Split data
  • using function 'split_data()' in the preare.py file included in the Github repository
  • data is split into train validate and test datasets using the 60,20,20 method.
• Explore the data using visualizations and stats tests.
• Using the best features run classification models on the train and validate to find the best model.
• Logistic Regression was the best/most consistent performing model.
• Read over the outputs and form a conclusion and summarize with recommendations/nextsteps.

Summary

• Completed the data analysis pipeline (Acquire, Prepare, Explore, Modeling)... twice.
• Selected features (strokes gained approach, off-the-tee, tee-to-green, total) were statistically significant, especially when using 2-week averages in the second iteration.
• Ran scaled data with selected features through four predictive models:
  • In iteration 1, logistic regression performed the best with 61% accuracy on unseen test data.
  • In iteration 2, the decision tree outperformed other models with 73% accuracy, surpassing the baseline and the iteration 1 model by 17% and 12%, respectively.

Recommendations

• Use the iteration 2 model as a guide for predicting next week's performance with 73% accuracy based on recent performance averages.
• Focus on improving the following factors to increase chances of making the cut:
  • Strokes gained approach
  • Strokes gained off-the-tee
  • Strokes gained tee-to-green
    • Tee ball accuracy and iron accuracy are crucial for minimizing strokes lost to the field.
• Caveat: It is impossible to predict with 100% certainty due to unquantifiable variables.

Next Steps

• Further investigation is required for a more robust/powerful predictive model:
  • Incorporate weather attributes through clustering.
  • Include course slope and length in the analysis.
  • Explore player demographics (age, height, weight).
  • Develop specific models based on course-specific requirements.