Target Selection
================

First, users have to open the application from `Geoscience ANALYST`_'s Python menu, as shown in :numref:`figure_python_menu`. The application is located in the ``Pro Geology`` section. Just click on the ``Target Selection`` button to open the application.

.. _figure_python_menu:

.. figure:: ./images/target_selection/python_menu.png
    :align: center
    :scale: 80%

    *Open the Targeting workflow from Geoscience ANALYST's Python menu.*

When launching the application, a window appears, as shown in :numref:`figure_uijson`, prompting users to select the "target". These targets represent the mineralized points that will be used to identify similar points with potential mineralization. The target's value can be set using a range slider for continuous values or categories for referenced values. All points outside this range will be categorized as "negative" points (points with no known mineralization). All other numeric data layers (float, integer, and referenced) associated with the target will be employed in subsequent steps. Users can press ``OK`` to initiate the workflow.

.. _figure_uijson:

.. figure:: ./images/target_selection/target_selection.png
    :align: center
    :scale: 70%

    *Select the object containing the data, and define the target to use.*

.. _Geoscience ANALYST: https://www.mirageoscience.com/mining-industry-software/geoscience-analyst/

As previously noted, this process involves searching for statistical correlations between mineralization and various properties, requiring a statistically representative number of positive points—ideally at least a few hundred. Operating the application with fewer points can lead to unreliable results, not only because the limited data may fail to capture the necessary correlations for accurate analysis, but also because it might suggest spurious correlations that don't actually exist due to the insufficient sample size.

Optionally, negative points can be chosen. However, selecting negative examples carries significant statistical implications. Instead of identifying "anomalous" points against a background, the application will assign a probability for every point to belong to either the positive or negative class. Thus, a point not sharing properties with either class could be classified as "positive", regardless of its mineralization probability.

Geologists often worry that positive points not mapped might be included in the negative dataset. However, since positive points are generally scarce and the algorithm seeks statistical trends rather than case-by-case scenarios (unless it suffers from `overfitting`_, then unreliable), having a few positive points in the negative dataset should not be problematic.

One scenario where using negatives is relevant is when a significant proportion of positive points is expected in the entire dataset (>20%), which is exceedingly rare in mineral prospectivity.

.. _overfitting: https://aws.amazon.com/what-is/overfitting/