Scholarlink

<p>Maritime transportation mode use pattern differs from land transportation. However, few have investigated passenger ferry mode choice. Therefore, this study aims to investigate what drives passengers to use ferry. This study employed Artificial Neural Networks (ANN) with 4 different activation functions: ANN identity, ANN logistic, ANN tanh, and ANN ReLU. A stated preference survey regarding Langsa-Penang passenger ferry route was conducted in Langsa, Aceh, Indonesia in 2019 which returned 4020 observations of existing and hypothetical scenarios that contained 12 variables of trip attributes and socio-demographic characteristics. ANN was used because of its capabilities to relax underlying assumptions often employed in statistical analysis and capture complex relationship among the data. To compensate for the lack of interpretability often associated with ANN, variable impacts on travel mode choice models were investigated and their importance were ranked. The calibrated ANN models were tested to forecast mode choice to investigate the generalizability of those models. To prevent bias, 80:20 training-testing split, and 10-fold cross-validation were used. ANN models’ predictive capabilities were compared to Logit model as a benchmark. This study finds that ANN models are capable of forecasting travel mode choice more accurately and also exceeds the predictive performance of the logit model. ANN tanh accuracy (87%) was almost 9% more than logit’s. ANNs’ performance was also observed to exceed logit’s in forecasting travel mode choice with limited observation. However, ANN models were time-consuming. Furthermore, each variable has varying significance in different models. The best-performing functions, ANN tanh and ANN ReLU, have the number of family members and dummy young age as the most significant variables. Overall, the mean of all model rankings indicated that the dummy middle-high transport expenditure was the most crucial variable while trip frequency was the least significant. By having accurate models, the findings from this study can be used for better understanding and effective modeling of passenger ferry travel choice behavior.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Application of Artificial Neural Networks (ANN) in Investigating Travel Mode Choice: A Case Study of Langsa-Penang Passenger Ferry

  • 期刊论文

  • 发表日期：

    2025年6月6日

查看全文
  • Fadhlullah Apriandy,
  • Sugiarto Sugiarto,
  • Sofyan M. Saleh,
  • Lulusi Lulusi

摘要

Maritime transportation mode use pattern differs from land transportation. However, few have investigated passenger ferry mode choice. Therefore, this study aims to investigate what drives passengers to use ferry. This study employed Artificial Neural Networks (ANN) with 4 different activation functions: ANN identity, ANN logistic, ANN tanh, and ANN ReLU. A stated preference survey regarding Langsa-Penang passenger ferry route was conducted in Langsa, Aceh, Indonesia in 2019 which returned 4020 observations of existing and hypothetical scenarios that contained 12 variables of trip attributes and socio-demographic characteristics. ANN was used because of its capabilities to relax underlying assumptions often employed in statistical analysis and capture complex relationship among the data. To compensate for the lack of interpretability often associated with ANN, variable impacts on travel mode choice models were investigated and their importance were ranked. The calibrated ANN models were tested to forecast mode choice to investigate the generalizability of those models. To prevent bias, 80:20 training-testing split, and 10-fold cross-validation were used. ANN models’ predictive capabilities were compared to Logit model as a benchmark. This study finds that ANN models are capable of forecasting travel mode choice more accurately and also exceeds the predictive performance of the logit model. ANN tanh accuracy (87%) was almost 9% more than logit’s. ANNs’ performance was also observed to exceed logit’s in forecasting travel mode choice with limited observation. However, ANN models were time-consuming. Furthermore, each variable has varying significance in different models. The best-performing functions, ANN tanh and ANN ReLU, have the number of family members and dummy young age as the most significant variables. Overall, the mean of all model rankings indicated that the dummy middle-high transport expenditure was the most crucial variable while trip frequency was the least significant. By having accurate models, the findings from this study can be used for better understanding and effective modeling of passenger ferry travel choice behavior.