The predictive accuracy of Machine Learning (ML) and Artificial Intelligence (AI) models has increasingly encouraged their application in the pro-perty evaluation studies and, more generally, in the economic research. While the predictive potential of these models is now widely acknowledged, they have often been criticised for being likened to ‘black boxes’ that yield results that are difficult to interpret. In this regard, thanks to some recent advances in the field of eXplainable Artificial Intellicence (XAI), there is a growing debate on the need to use explainability techniques to make the predictions of ML models easier to interpret and, consequently, more reliable. This paper aims at a twofold objective. Firstly, it intends to investigate the importance of explainability models for a transparent analysis of the predictions returned by ML and AI. Secondly, it proposes a novel and automated methodological approach that integrates Artificial Neural Network (ANN) with a Geographic Information System (GIS). The use of GIS has the advantage of geo-localising on a map both the input data, i.e. the intrinsic and extrinsic characteristics of each property, and the results of the model, expressed in terms of expected prices. In addition, the method-logical approach envisages the use of an Explainer Model (EM) that allows for a transparent interpretation of the ANN results, making it possible to read the contribution of each variable to the expected price. In summary, the methodology thus defined is able to provide planners and decision-makers with a completer and more reliable picture of real estate trends. Case study applications will test the proposed ANN-GIS approach.
An Artificial Neural Network Based Model for Urban Residential Property Price Forecasting
Gabrielli L.;
2024
Abstract
The predictive accuracy of Machine Learning (ML) and Artificial Intelligence (AI) models has increasingly encouraged their application in the pro-perty evaluation studies and, more generally, in the economic research. While the predictive potential of these models is now widely acknowledged, they have often been criticised for being likened to ‘black boxes’ that yield results that are difficult to interpret. In this regard, thanks to some recent advances in the field of eXplainable Artificial Intellicence (XAI), there is a growing debate on the need to use explainability techniques to make the predictions of ML models easier to interpret and, consequently, more reliable. This paper aims at a twofold objective. Firstly, it intends to investigate the importance of explainability models for a transparent analysis of the predictions returned by ML and AI. Secondly, it proposes a novel and automated methodological approach that integrates Artificial Neural Network (ANN) with a Geographic Information System (GIS). The use of GIS has the advantage of geo-localising on a map both the input data, i.e. the intrinsic and extrinsic characteristics of each property, and the results of the model, expressed in terms of expected prices. In addition, the method-logical approach envisages the use of an Explainer Model (EM) that allows for a transparent interpretation of the ANN results, making it possible to read the contribution of each variable to the expected price. In summary, the methodology thus defined is able to provide planners and decision-makers with a completer and more reliable picture of real estate trends. Case study applications will test the proposed ANN-GIS approach.I documenti in SFERA sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.