ESTIMASI TINGGI BANGUNAN MENGGUNAKAN MACHINE LEARNING UNTUK PEMODELAN PETA HISTORIS 3D

• Main Author: Nisya Fitri, An
• Format: Final Project
• Language: Indonesia
• Subjects: Teknik (Rekayasa, enjinering dan kegiatan berkaitan)
• Online Access: https://digilib.itb.ac.id/gdl/view/61443
• Institution: Institut Teknologi Bandung

Map visualization in 3D is gaining increased popularity as its potential is increasing. A 3D Model can be used in the spatial aspect analysis for urban city planning and construction as a basis for policy making, urban environment studies, historical documentation, and other applications. The increase of 3D model potential is followed by an increase in variation and quality of data acquisition, such as terrestrial mapping, aerial photogrammetry, and satellite imagery. The data available in the past is not as comprehensive as the data available in the present due to limitations in data acquisition methods in the past as opposed to the present. Historical maps are examples of data available from the past. These maps are commonly found in physical form, have small scales, and do not include the height of objects depicted in the map. A computer is believed to be able to learn patterns in data, then form a model from the aforementioned patterns. This model may be used to answer similar problems from other datasets. This ability of a computer is known as machine learning, where an algorithm is formed using a programming language to form a data-based model. There are a variety of machine learning algorithms and models which may be used accordingly to classify or predict data. Building height in the past may be estimated using a machine learning regression model by first studying the pattern of similar data in the present which accommodate height data as a target variable and building characteristics as predictor variables. In this undergraduate thesis, building height estimation in 1940 is done at Kawasan Keraton Cirebon using a machine learning program, specifically the random forest regression model by learning the patterns of building data in 2020 resulting R-Squared value 61.2%. Various metrics for estimated building height are calculated using MAE, MSE, and RMSE as 1.12, 1.75, and 1.32 meters respectively. The result of building height estimation in 1940 is then visualized in a 3D LoD 1 model along with the 3D model in 2020.