유동 인구 분석을 위한 딥러닝 기반 알고리즘 연구 = A Study on Deep Learning-Based Algorithms for Population Flow Analysis

다국어 초록 (Multilingual Abstract)

Purpose The purpose of this study is to integrate various external data sources-including transportation, meteorological, credit card consumption, and demographic data-to precisely analyze fluctuations in tourist population flows in regional attractions and to verify the predictive capability of deep learning algorithms (LSTM). This study aims to complement the limitations of traditional statistical approaches and provide empirical evidence that can contribute to tourism demand forecasting and local commercial district management.


Design/Methodology/Approach As a research method, a time-series dataset was constructed by integrating transportation and credit card consumption data from the Busan region. During the preprocessing stage, missing values were interpolated, categorical variables were label-encoded, and derived features such as holiday flags and moving averages were generated. Subsequently, categorical variables were processed through embedding layers, while continuous variables were standardized and used as model inputs. Finally, a BiLSTM-based prediction model was designed.


Findings Experimental results demonstrated that the proposed model achieved an MAE of 0.26 and a MAPE of 15.3% on the test dataset. These findings suggest that deep learning-based time-series models can more effectively capture changes in population flows compared to conventional statistical methods.

다국어 초록 (Multilingual Abstract)

Purpose The purpose of this study is to integrate various external data sources-including transportation, meteorological, credit card consumption, and demographic data-to precisely analyze fluctuations in tourist population flows in regional attractions and to verify the predictive capability of deep learning algorithms (LSTM). This study aims to complement the limitations of traditional statistical approaches and provide empirical evidence that can contribute to tourism demand forecasting and local commercial district management.




Design/Methodology/Approach As a research method, a time-series dataset was constructed by integrating transportation and credit card consumption data from the Busan region. During the preprocessing stage, missing values were interpolated, categorical variables were label-encoded, and derived features such as holiday flags and moving averages were generated. Subsequently, categorical variables were processed through embedding layers, while continuous variables were standardized and used as model inputs. Finally, a BiLSTM-based prediction model was designed.




Findings Experimental results demonstrated that the proposed model achieved an MAE of 0.26 and a MAPE of 15.3% on the test dataset. These findings suggest that deep learning-based time-series models can more effectively capture changes in population flows compared to conventional statistical methods.