stoae / abductiu

Teleològic i abductiu: Una aproximació als models additius per la interpretabilitat.

This repository contains open-source code for the work: Teleològic i abductiu: Una aproximació als models additius per la interpretabilitat.

Abstract

This work analyzes the syntax of models and, through the methodology of data science, the ability to predict and interpret the models with a geolocated open data set of the housing market. The aim is twofold a) to get answers about the predictive abilities of explanatory models b) From a machine learning perspective, to look for the best model to interpret the data. Six different models of algorithms are trained, which are then used for both prediction and interpretability. The best explanatory model for predictions is Spline and the best model with interpretability is Neural Additive Models. The findings underline the fact that explanatory models achieve similar results to non-interpretive ones. It also emphasizes the importance of models with interpretability by design while addressing the challenges posed by the creation of reliable systems.

Teleològic

Abductiu

Models

Exploratori

Modelatge

Visualització

Resultats Predictius Models