| Popis: |
Tema rada je „Geografski ponderirana neuronska mreža“. U radu se naglašava problem prostorne heterogenosti te njegovo rješavanje. Analizira se geografski ponderirana regresija te se ukazuje na njezine nedostatke kod predviđanja heterogenog prostora. Predstavlja se i opisuje rješenje za geografski ponderiranu regresiju pomoću geografski ponderirane neuronske mreže koje su implementirane u programskom jeziku Python. Razlika između geografski ponderirane regresije i geografski ponderirane neuronske mreže prikazana je njihovim testiranjem na četiri skupa umjetnih podataka različite prostorne heterogenosti i vizualizacijom njihovih rezultata. The topic of the paper is "Geographically weighted neural network". The paper emphasizes the problem of spatial heterogeneity and its solution. Geographically weighted regression is analyzed and its shortcomings in predicting heterogeneous space are pointed out. A solution for geographically weighted regression using geographically weighted neural networks implemented in the Python programming language is presented and described. The difference between geographically weighted regression and geographically weighted neural network is shown by testing them on four sets of artificial data of different spatial heterogeneity and visualizing their results. |
