The Arc is a small Streamlit app that predicts an NBA player’s Year 5 points per game (PPG) using their Year 2 “sophomore season” stats.

The idea: the jump from Year 1 → Year 2 contains strong signals about a player’s long‑term ceiling. This app turns those “sophomore signals” into a simple Year 5 projection.

• Lets you search for an NBA player and:
  • See their Year 2 stats (PPG, RPG, APG, efficiency, minutes, etc.).
  • Get Year 5 PPG predictions from two different modeling paths.
  • Compare those predictions to the actual Year 5 PPG (when available).
• Groups players into data‑driven archetypes (via clustering) so you can see what “type” of player they are.
• Shows model performance:
  • Error distributions for each path.
  • Feature importance to see which stats drive the predictions.
  • Best and worst individual predictions.

The project runs two competing modeling strategies:

• Uses a small, standard feature set:
  • Year 2 box score stats (points, rebounds, assists, minutes, shooting splits).
  • Simple growth metrics (deltas from Year 1).
  • Basic context (draft position, AST/TOV).
• Trains a regression model with default hyperparameters.
• Goal: fast, interpretable, “good enough” baseline.

• Uses all Path 1 features plus engineered features, such as:
  • Skill Diversity Index (improvement across multiple categories).
  • Usage‑to‑Efficiency ratio.
  • Draft overperformance.
  • Minutes trajectory.
  • Free‑throw improvement.
• Adds hyperparameter tuning to squeeze out extra accuracy.
• Goal: maximum accuracy and richer basketball intuition.

• Home
  High‑level project overview and quick comparison of Path 1 vs Path 2 performance.

• Scouting Report

  • Select a player and see:
    • Year 2 stats.
    • Dual Path 1 vs Path 2 Year 5 predictions.
    • Actual Year 5 PPG and errors (when data exists).
  • Career trajectory chart and archetype radar.

• The DNA Explorer

  • View the 5 data‑driven archetypes.
  • See average stats per archetype.
  • Explore a 2D map of players colored by archetype.

• Model Analysis

  • Detailed methodology for both paths.
  • Side‑by‑side metrics (MAE, R², training time, overfitting checks).
  • Feature importance charts.
  • Error histograms and best/worst predictions.

• Python
• Streamlit for the web app UI
• pandas / NumPy for data handling
• scikit‑learn for modeling + clustering
• Plotly for visualizations