Data Narrative: Tennis Major Tournaments Dataset

This repo presents forth a Data Narrative prepared as a part of a course project. The data analysis involves the use of Clustering, Classfication, and Regression using machine learning models in python's Scikit-Learn Library. The repo also includes a written report for the same as compiled in LaTex.

The Overview

The Narrative was part of the ES-114: Probability, Statistics, and Data Visualization course offered to the first-year B.Tech students during the April '23 for the first time at the Indian Institute of Technology, Gandhinagar. The Machine Learning Course ES-113: Data Centric Computing here at IIT, Gandhinagar, was conducted in coherence with this course. The Narrative allowed for the analysis of the Tennis Major Tournaments dataset using Clustering, Classification, and Regression. Several of Probabilistic and Data Analystic tools like Covariance Matrices, Correlation Factors, and Principal Components Analysis were employed in the narrative.

Libraries used and Analysis Approach

The following is the list of Python Libraries used for the narrative:

1. Pandas
2. NumPy
3. Scikit-Learn

• sklearn.naive_bayes.GaussianNB
• sklearn.preprocessing.StandardScaler
• sklearn.decomposition.PCA
• sklearn.cluster.KMeans
• sklearn.metrics.accuracy_score
• sklearn.model_selection.train_test_split
• sklearn.model_selection.cross_val_score

4. Plotly

• plotly.express.scatter()
• plotly.express.box()
• plotly.express.3d_scatter()
• plotly.express.bar()

• The main aim of the analysis was to predict the target (win: 0/1) of the newer result based upon the features given to us to work with.
• To prepare our data for training, number of features had to be reduced down.
• Correlation among the features was analysed and redundant features having high correlation were shot-down.
• Now with the less features, the data was prepared for training on a Gaussian Naive Bayes classifier.
• Accuracy of this model was tested using a 5-fold cross validation score. The model actually gave a high accuracy score of 0.9049230769230769.
• Another way of unsupervised learning was the use of KMeans clustering.
• Two clusters were allowed to form and the accuracy was compared with original data.
• To visualize the clustering, the dimensionality of the features was reduced to 3 using Principal Component Analysis (PCA) so clusters were plotted in 3D.
• Finally, there were some ways using Pandas and NumPy to filter out important statistical data/stats and the same was plotted for using Plotly graphs.

Please refer to the Colab link to view the interactive Plotly graphs.