Refactor predict() to use grouped iteration instead of O(n²) dataframe filtering

[midaa1]

Summary

This PR optimizes the data aggregation step inside predict() in this issue #74 by removing row-wise dataframe iteration and repeated filtering.

Changes Made

Replaced:

• df_data.iterrows() loop

• Per-row dataframe filtering based on (True X, True Y)

Added:

• Single-pass iteration over:

• df_data.groupby("True XY")

• Direct extraction of grouped predictions using:

group["Predicted X"].tolist()
group["Predicted Y"].tolist()

• Reused existing precomputed metrics:

   - precision_xy
  
   - accuracy_xy
  

Before

• Nested logic caused repeated full-dataframe scans.

• Complexity: O(n²).

• Duplicate processing for identical truth coordinates.

After

• Each (True X, True Y) group processed once.

• Complexity reduced to O(n).

• Cleaner and more readable logic.

• Performance Impact

• Significant speed improvements on larger datasets.

• Reduced CPU usage and memory overhead.

Behavior Verification

• Output dictionary structure remains unchanged.

• Precision and accuracy values unchanged.

• Filtering of negative predictions still respected.

Notes

This refactor prepares the code for better scalability and aligns the aggregation logic with pandas best practices.