DSpace Collection:http://hdl.handle.net/10203/254112024-03-29T14:34:16Z2024-03-29T14:34:16ZMachine learning study on real estate massive assessments: pseudo self comparison methodChoi, Seungwoohttp://hdl.handle.net/10203/3093092023-06-23T19:34:54Z2022-01-01T00:00:00ZTitle: Machine learning study on real estate massive assessments: pseudo self comparison method
Authors: Choi, Seungwoo
Abstract: however, one limitation of the model is that it does not consider the volatility of the real estate market.
To address this, we propose the pseudo self comparison method (PSCM). A pseudo self is defined as a real estate property with mostly the same hedonic features and similar price volatility as the target property. Examples include housing properties with different numbers of stories but the same floor plan, such as apartments and condominiums. The previous transaction price of the pseudo self and changes in the real estate market during the period of previous transaction were used as variables. To compare the proposed model with the HPM, we constructed a dataset of apartment transactions in Seoul and its surrounding region, Gyeonggi, and compared the price estimation results. The proposed technique was able to reduce the mean average percentage error by approximately one-fifth. Moreover, the error in price estimation did not significantly increase even when there was an increase in the volatility in the real estate market, demonstrating the model’s robustness.
Although the PSCM showed high estimation accuracy for real estate prices, it is difficult to apply it to multiplex houses and detached houses, whose hedonic features are distinctive. Therefore, we expanded this approach to develop a generalized PSCM. This method adds a module that searches for pseudo self candidates (PSCs) and a module that searches for similar transactions among the previous transactions of the PSCs. The first module applies a two-step clustering process; an initial cluster is formed based on the locational proximity context, after which a sub-cluster is formed within it based on the price context. For the second module, two previous transactions are matched based on the PSCs and then input into a deep learning-based similar transaction search module, after which the model is trained by predicting the dissimilarity score. Using the two proposed modules, the pseudo self’s previous transactions were selected even for real estate properties with completely different hedonic features, based on which PSCM variables were generated. The newly proposed approach outperformed the HPM in terms of price estimation ability, and the price estimation model combined with the HPM achieved an even higher performance. The results suggest that our proposed method adequately accounts for factors that are not considered in the existing HPM.
To conclude, the proposed PSCM regards property values in the same manner as real estate appraisers while also accounting for changes in the market; Volatility in real estate values greatly impacts the national economy and can also act as a warning signal for financial crises. As such, a wide range of studies have been conducted to accurately estimate the transaction prices of real estate properties. The hedonic price model (HPM) is a long-used methodology for estimating real estate values and is still extensively used across various fields today. It estimates the transaction price based on the structural characteristics and living conditions of the real estate property, as well as the characteristics of the surrounding environment. The HPM is advantageous for analyzing the relationship between the characteristics and price of a property; however, it also minimizes the subjective interference from the appraiser that is required for the actual evaluation and achieves high price estimation performance using only an automated approach. Therefore, the proposed PSCM is a suitable framework for mass valuation models of real estate properties.
Description: 학위논문(박사) - 한국과학기술원 : 지식서비스공학대학원, 2022.2,[vi, 96 p. :]2022-01-01T00:00:00ZNeural network modeling for rainfall prediction: observation extrapolation and numerical forecast correctionJeong, Chang-Hoohttp://hdl.handle.net/10203/3093082023-06-23T19:34:53Z2022-01-01T00:00:00ZTitle: Neural network modeling for rainfall prediction: observation extrapolation and numerical forecast correction
Authors: Jeong, Chang-Hoo
Abstract: Torrential rain and inundation are becoming more common because of global warming and extreme weather, leading to an increase in human and property damage. It is critical to improve the performance of the rainfall prediction model in order to minimize damage by identifying as early as possible when a natural disaster may occur. In this study, we propose a model that performs very short-term rainfall prediction based on past observation data and a model that performs short- and medium-term rainfall correction based on future forecast data, respectively, to improve the performance of rainfall prediction. In Study 1, we focused on the performance improvement of the nowcasting model, which predicts very short-term rainfall using radar extrapolation. We improved radar extrapolation performance by extending the architecture of the existing encoding-forecasting model to emphasize the phenomenon of the most recent data. However, the nowcasting method based on spatiotemporal analysis of radar data is limited in that the prediction frame blurs as the time step increases, even though it accurately predicts near future rainfall. Furthermore, there is a limitation in that it does not properly represent the sudden change in rainfall patterns caused by rapid changes in weather conditions. Therefore, Study 1 is a method suitable for very short-term rainfall prediction, and a new method needs to be developed for short- and medium-term rainfall prediction. In Study 2, we proposed a method to improve the accuracy of forecast data by correcting future simulation data produced by a numerical weather prediction (NWP) model close to the ground truth data observed, as a method for predicting short- and medium-term rainfall. The NWP model is the pinnacle of human knowledge about the Earth's atmospheric circulation, currently used for weather forecasting in the practical field, however, room for improvement remains because there are many differences from the actual ground truth. Therefore, Study 2 proposed a deep learning-based correction method that transforms the distribution of the forecast data of the NWP model into the distribution of the ground truth data measured by ground observation equipment such as radar by applying the GAN model, which exhibits excellent performance in data distribution transformation in different domains. Using this method, the quality of the forecast data is improved by correcting the error of the simulation result of the NWP model close to the ground truth. In Future Study, we will develop an integrated rainfall prediction model that combines the extrapolation result of radar data from Study 1 and the correction result of forecast data from Study 2 to produce the final rainfall prediction result. By combining the two results, the final model can respond to rapid changes in future weather conditions while mitigating the spatial smoothing problem. In addition, through the future study, we can improve the rainfall prediction model's performance by combining the accuracy of the prediction data obtained in Study 1 and the stability of the forecast data obtained in Study 2. Overall, this study dealt with rainfall prediction, which is a field of weather forecasting, but it also presents practical guidelines for the design of integrated deep learning models for various weather forecasting fields that fuse past observation data and future forecast data using spatiotemporal prediction modeling.
Description: 학위논문(박사) - 한국과학기술원 : 지식서비스공학대학원, 2022.2,[iv, 82 p. :]2022-01-01T00:00:00Z(A) study on big data-based building life-span prediction method for life cycle assessment and life cycle cost using machine learningJi, Sukwonhttp://hdl.handle.net/10203/3093122023-06-23T19:34:55Z2022-01-01T00:00:00ZTitle: (A) study on big data-based building life-span prediction method for life cycle assessment and life cycle cost using machine learning
Authors: Ji, Sukwon
Abstract: Various methods are used to make major decisions in the construction industry. Among them, life cycle assessment (LCA) method and life cycle cost (LCC) analysis are mainly used for environmental and economic evaluation of buildings, construction methods, and materials. To perform these methods properly, it is essential to estimate the realistic lifespan of the building. However, since it is practically impossible to estimate the lifespan by considering various factors affecting the lifespan of a building, most LCA and LCC studies have assumed the building lifespan uniformly according to the major structural types of buildings. However, the lifespan of buildings is in fact very variable, and a simple assumption that all buildings of the same structural type follow the same lifespan may lead to completely erroneous analysis of the results of LCA and LCC studies. In the first study, 1,812,700 records of buildings constructed and demolished in South Korea were collected, the actual lifespan of each building was analyzed, and a building lifespan prediction model using deep learning and traditional machine learning was developed. As a result, in the case of a reinforced concrete building, which is generally known to have a durability of about 50 years, the actual average building lifespan was only 22.8 years. In addition, the prediction model
investigated in this study showed root mean square error (RMSE) of 3.72~4.6 and coefficient of determination of 0.932~0.955. Among them, the deep learning-based prediction model was found to have the best performance. Therefore, these results mean that realistic LCA and LCC analysis results cannot be obtained by simply estimating the lifespan of a building in the existing method of determining the life cycle of a building with several specific factors. The second study empirically demonstrates the effect of the application of big data-based realistically predicted lifespan on LCA and LCC analysis via LCA and LCC analysis on waterproofing methods of building models and actual buildings. In the LCA and LCC analysis of the waterproofing method of the architectural model, the application of the building lifespan from 5 to 39 years decreased carbon emissions by a maximum of 78% to a minimum of 24% in all phases relative to the result when a building lifespan of 50 years was assumed. During the maintenance phase, the maximum and minimum reductions were 100% and 31 %, respectively. As a result of the LCA analysis of the waterproofing method of 17 real building cases, the accuracy of the LCA analysis results
revealed a tendency to be irregular as the predicted lifespan percent error increased. Moreover, the percent error of the LCA analysis results of the buildings with a predicted life percent error of 6% or less was 0%. Evidently, depending on the research purpose, a predictive building lifespan model that can guarantee a certain percent error or less is required for accurate LCA results.
Based on the above research results, it is evident that the building life prediction method that applies deep learning based on big data is the most realistic building life prediction method thus far, and it is necessary to apply an accurately predicted lifespan to secure the accuracy of building-related LCA and LCC analysis. Therefore, this study demonstrates that a big data-based building lifespan prediction method is an essential and promising direction for effectively guiding business planning and critical decision-making throughout the construction process.
Description: 학위논문(박사) - 한국과학기술원 : 지식서비스공학대학원, 2022.8,[vi, 105 p. :]2022-01-01T00:00:00ZMeta-learning for recommender systemsKim, Minseokhttp://hdl.handle.net/10203/3093132023-06-23T19:34:55Z2022-01-01T00:00:00ZTitle: Meta-learning for recommender systems
Authors: Kim, Minseok
Abstract: The widespread of mobile devices has enabled people to use various online services, such as video streaming, shopping, and news in everyday life. Because users seek only a few items out of a myriad of items in such services, the role of a recommender system is to quickly find them out to enhance user satisfaction and help the growth of the service. However, there are three challenges in recommender systems due to the vulnerability of the standard learning scheme: low-quality training data, online update delay, and unfair recommendations after learning.This dissertation aims to resolve the three challenges of recommender systems via meta-learning, which is also known as “learning to learn.” Meta-learning can deal with the vulnerability of recommender learning by adaptively guiding its learning process after observing how the recommender learns. In this regard, this dissertation suggests meta-learning-based approaches to overcome the three challenges.The first study proposes PREMERE that performs training data reweighting to avoid learning from low-quality data. PREMERE adaptively provides a high weight on useful but sparse data, and a low weight on missing data to induce the recommender to learn only from qualified data. Through extensive experiments on three real-world benchmark recommender datasets, PREMERE proved its effectiveness, improving performance by up to 26.9% compared with state-of-the-art algorithms. The second study proposes MeLON that online updates recommender systems with up-to-date information. While current user interests change constantly, their sparse signal in the user-item domain makes recommenders difficult to catch up due to the fixed scheme of standard learning. To this end, MeLON adaptively generates learning rates between each data and model parameter considering theirmutual relationship. Extensive empirical evaluations on three real-world online service datasets validated that MeLON can continuously maintain the high performance of a recommender system during online service by virtue of its high flexibility for updates.The third study first performs data analysis to demonstrate that recommender systems are prone to over-recommend popular items especially to new users, which is undesirable for both users and item providers. Because it is fair to recommend popular items to a user according to their ratio in the user’s history, to cope with this challenge, the third study proposes ColA that expedites learning of tail-itemsadapted to each user and recommender model learning state. Experimental results on a real-world recommender system dataset confirm that ColA can achieve higher performance and improved fairness compared with state-of-the-art popularity debiasing algorithms.This dissertation is expected to enhance user satisfaction, the ultimate goal of recommender systems by effectively addressing the challenges caused by learning vulnerability.
Description: 학위논문(박사) - 한국과학기술원 : 지식서비스공학대학원, 2022.8,[v, 61 p. :]2022-01-01T00:00:00Z