Árbol de Decisión de Ruta Óptima

Abstract

This thesis presents an Optimal Path Decision Tree (OPDT), an innovative regression-based decision tree algorithm featuring a specialized split function designed to simplify the creation of decision trees and enhance the comprehension of decision paths. This enhancement is particularly beneficial for their application in framing decision-making policies. In contrast to traditional approaches, the OPDT algorithm singularly utilizes predictive variables to determine the most efficacious pathway to reach specific objectives, thereby minimizing the complexity and bolstering the interpretability of the generated trees.

The versatility of the OPDT model was thoroughly investigated through its application to both regression and classification datasets, thereby emphasizing its adaptability. The effectiveness of the model was validated through empirical analyses using datasets sourced from public repositories, thereby demonstrating its applicability across diverse fields. A comparative evaluation of the established Classification and Regression Trees using the Gini index (CART-Gini) method highlights the unique approach of the OPDT model. The Gini index optimizes thresholds for sample division to improve predictability, and the OPDT identifies a singular optimal path, enhancing the probability of meeting a specific target. OPDT represents a specific functionality within the CART algorithm family. This makes OPDT particularly adept within the CART-Gini family for tailoring decision policies and effectively bridging the gap in the capabilities of the original CART-Gini framework.

Additionally, this thesis delves into incorporating OPDT logic into ensemble methodologies, such as Bagged Trees and Random Forests, suggesting a broader applicability of OPDT principles. This thesis concludes that the OPDT model represents the most concise decision tree approach for achieving targeted outcomes. The algorithm and code developed in this study contribute significantly to the knowledge of machine learning in the context of decision trees and their integration into decision-policy frameworks.