This project aims to predict the career longevity of NBA players based on their performance statistics. The objective is to predict whether a player will have a career that lasts more than 5 years (Class 1) or less than 5 years (Class 0). The project includes two Jupyter notebooks:
- EDA (Exploratory Data Analysis) Notebook: Contains all the steps of data analysis, cleaning, and feature engineering to prepare the dataset for modeling.
- Modeling Notebook: Contains a list of models tested, including the best-performing model, which generalizes the best for the prediction task.
After testing different models, SVM was selected as the best model for predicting whether a player will have a career lasting more than 5 years or less than 5 years. The best parameters for the SVM model are:
C: 0.1gamma: 'scale'kernel: 'linear'
The SVM model was chosen based on its superior performance in key metrics compared to other models like AdaBoost and Logistic Regression. Specifically:
Accuracy: The SVM model achieved an accuracy of 73%, slightly outperforming AdaBoost at 71%. This suggests that SVM is more reliable in making overall correct predictions.
Recall: SVM demonstrated a recall of 0.86 for class 1, indicating it was more effective in identifying long-career players.
A Streamlit app has been developed to allow users to input data and predict whether an NBA player will have a long or short career. You can access the app at: