The team consists of the following 4 members: Alex Glickman, Ethan Jones, Matt Ansaldi and Cecily Martin
An Olympic data analysis is being conducted on 120 years of Olympic history. Specifically, data on athletes and medal results will be analyzed from Athens 1896 to Rio 2016 to explore whether a country’s GDP, population and team size have an impact on a country’s Olympic medal counts.
We're all interested in sports.
- Data Sources:
- 120 Years of Olympic History: Athletes and Results (Datasets: athlete_results.csv and noc_regions.csv )
- Olympic Games Hosts (Dataset: olympic_hosts.csv)
- GDP per Capita for Countries by Year
- Population for Countries by Year
- Latitude and Longitude for Every Country and State (Dataset: world_country_and_usa_states_latitude_and_longitude_values.csv)
- Entity Relationship Diagram (ERD) Tool: Quick Database Diagrams
- Languages: Python
- Python Dependencies: pymongo, scikit-learn, pandas, numpy, seaborn, matplotlib, SciPy
- Applications: MongoDB, Tableau
- Tools: MS Excel, Jupyter Notebook
Data sources utilized were found via Kaggle.com and Gapminder.org. Gapminder combines data from a variety of sources.
- 120 Years of Olympic History: Athletes and Results dataset (athlete_events.csv and noc_regions.csv) were found on Kaggle, but the source of the data stems from www.sports-reference.com
- Olympic Games Hosts dataset (Olympic_hosts.csv) was found on Kaggle, but the source of the data stems from www.olympic.org
- GDP dataset was found on GapMinder, but the source of the data stems from The World Bank World Development Indicators, Maddison and The IMF source 1 and source 2.
- Population dataset was found on GapMinder, but the source of the data stems from the UN-Population Division, World Population Prospects 2019 and Maddison
- Latitude and Longitude for Every Country and State dataset (world_country_and_usa_states_latitude_and_longitude_values.csv) was found on Kaggle and for the project’s analysis, the file was renamed to coordinates.csv in order to condense the naming convention. The data stems from Google’s Dataset Publishing Language
- Is there a correlation between GDP and medal count?
- How important is the country’s population in terms of winning medals?
- Which parameter plays the biggest part when it is tested on its own?
- Does a country’s change in GDP over time, have an impact?
- Which countries won the most medals over time?
- How many medals are forecasted to be won at upcoming Games (Tokyo 2020 / Beijing 2022)?
- What are the GDP forecasts for years with upcoming Games (Tokyo 2020 / Beijing 2022)?
- Which countries are expected to lead the medal tally?
- Do Olympic host countries win more medals?
- Does seasonality (winter / summer) have an effect?
- What is the mean age of athletes?
- Does athlete age / gender have an effect?
- Are there other factors that could impact Olympic medal counts?
- How have the Olympic Games evolved over time by looking at the participation of different nations, sports and events?
- A Trello board has been setup, with alerts for all deliverables. The board will be updated on an ongoing basis.
- Team communicates daily via Slack
- Team meets via Google Meet, outside of scheduled 2-hour class sessions, twice per week and attends office hours. See calendar.
- A Google Drive is utilized for storing meeting notes, datasets and other files which enable collaboration.
- Each team member has an individual branch on the GitHub project repository.
A provisional database was created in MongoDB. Lines 3 - 5 within the cleaning_machine_learning.ipynb Jupyter Notebook connect to the database, upload data to the database and shows data being taken off of the database.
This ERD highlights the flow of information from one table, or CSV file, to another and captures the primary keys, foreign keys, and data types for each column, within the corresponding CSV file.
Pandas will be used to clean the data and perform an exploratory analysis.
Before the data was fed into the machine learning model, it was preprocessed. The first- and most-time consuming step was to merge the datasets. The most common column to merge on was the column containing the country name. This is where most of the cleaning was done as names such as St. Lucia versus Saint Lucia wouldn't be found as a match. The majority of this type of cleaning was done manually in Excel since writing lines of code for each country name that didn't match would've been even more time consuming.
Once the cleaning was done, it was time to add the code for the machine learning model. The machine learning model we chose was random forest. This decision came from the fact that our data is labeled (points to supervised machine learning) and that our output (medal count) is numerical. Random forests are made up of many decision trees (a decision tree can be thought of as a collection of if else statements). Each tree in the random forest makes its own prediction on the output. After this, the average of all the outputs is computed to give you the final result. That's to put it in simple terms!
According to the regressor.score function, our model has about 80% accuracy. The RMSE (root mean square error) is about 5. This means that the standard deviation of the residuals (prediction errors) is about 5 medals. Residuals measure how far from the regression line the actual data points are. That's pretty good for about a weeks’ worth of work.
See cleaning_machine_learningV6.ipynb Jupyter Notebook.
The largest question that was answered is how correlated a country's total GDP is to its medal count. According to analysis, it does not have much of an affect on the medal count. In fact, there was generally a low correlation between most of the data versus the total medals. The highest correlation was team size, which makes sense do to have more of a chance to win with more players. That correlation is .83, so other factors come into play (like talent, for example). In all, not one piece of data truly seemed to affect the total medal count.
If this research were to continue, more machine learning models should be tested. The random forest had did have an 85% accuracy. More investigation, however, may see higher accuracy. Some models may be more complex and need more resources and data. With the time and resources used, random forest was used from the beginning and was not changed.
A few items the team would have wanted to expand on are:
- Analyze gender more
- Seasonality in terms of breaking out the data versus breaking out the data by year
- How a country performs for host countries vs non-host countries
The above are pieces of data the team either wanted to include or go into more indepth with.
Tableau was utilized to create an interactive dashboard to better visualize whether a participating and/or host country’s GDP per Capita, Population and Team Size has an impact on Olympic medal counts. The dashboard consists of a variety of charts to visualize the data, by using a map, pie charts, a bar chart, and numerous scatter and dot plots. The interactive elements consist of universal filters by Olympic Game year and adjustable zoom on the map. To view the dashboard via Tableau here.
Overtime 194 countries have participated in the Olympic Games, during the 120-year period of Athens 1896 – Rio 2016, however the number of participating countries varies by Olympic year. To drill down on the data by Olympic year, one can click on the Olympic logos. From 1924 – 1992 the summer and winter games were held in the same year, so the data filters accordingly, by year.
When looking at the map, one can tell if a country participated in the selected year, based off whether the country is shaded. The darker the shade of the country, the higher the GDP per Capita. If one hovers over the pie chart in the selected country, one can also see the medal breakdown or refer to the bar chart on the right to view the top winning countries in that select year. As can be seen from the unfiltered data, the US has consistently won the most medals overtime.
When looking at the total medal count distribution in the various scatter plots, one can see that population and team size are more correlated than GDP per Capita, as the data is less scattered.
In total, 22 countries have hosted an Olympic Games. Countries that host an Olympic game tend to win more medals than when they are a participating nation. To view a breakdown of how all participating teams performed by Olympic Game year, the Overall Team Performance plot visualizes the total medal count by team and year.
A presentation has been drafted, outlining the project. The presentation can be viewed in Google Slides.