Travel Mode Choice Prediction Using Deep Neural Networks With Entity Embeddings

Ma, Yixuan; Zhang, Zhenji

doi:10.1109/access.2020.2985542

Cited by 25 publications

(10 citation statements)

References 44 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For instance, Yixuan Ma et al. [17] combinate entity embedding with Deep Neural Network (DNN) to deal with the representation of category variable to improve the prediction accuracy, and the model achieved 88% prediction accuracy on the dataset London Travel Demand Survey. What is more, some of them try to combine with the tradition discrete model with the deep learning model.…”

Section: Literature Reviewmentioning

confidence: 99%

“…Traditional methods infer travel modes of the individuals based on utility theory which require the detailed characteristics of the travelers [2][3][4][5][6]. Recenty, the supervised learning methods arose much interest which have been widely applied to infer travel modes due to the outstanding pattern classsification ability [7][8][9][10][11][12][13][14][15][16][17][18][19]. But most of existing methods focus on inferring travel mode of a trip, little work is done on predicting next travel mode choice of individual.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Travel mode choice prediction based on personalized recommendation model

Lai

Wang

Zheng

et al. 2022

IET Intelligent Trans Sys

View full text Add to dashboard Cite

The accurate prediction of individual travel mode in the city is necessary for the development of urban traffic intelligence. Through deep analysis on distribution of travel modes, policy makers can make proper policies to improve traffic conditions. With the development of sensing techniques and telecommunication technologies, it is easy to collect travel data of individuals and use some popular models to detect travel modes. However, existing methods are limited to aggregation analysis and these models suggest that travelers' preference mainly depend on the characteristics of travelers, little is known considering to extract individual travel preference from individual historical trips. Due to collect the user information is expensive and time‐consuming, one of the difficult tasks facing the prediction of travel mode by traditional methods is the lack of user's profile. What is more, data of different travel modes are always imbalanced. Accordingly, the paper proposes a model Balance Multi Travel Mode Deep Learning Prediction (BMTM‐DLP) for individual travel mode prediction inspired by the idea of recommender system. For the first time, the theory of recommender system is introduced into the field of individual travel mode choice prediction and we take advantage of the model to extract individual travel preference for travel mode prediction. Further, the focal loss function module is introduced into the model to reduce the impact of unbalanced categories. Experiments show that the model performs well on both datasets LPMC and XSCR, achieving 92.2% and 82.8% prediction accuracy in respective. It showed that the proposed model performed well in individual travel mode prediction.

show abstract

Section: Literature Reviewmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Travel mode choice prediction based on personalized recommendation model

Lai

Wang

Zheng

et al. 2022

IET Intelligent Trans Sys

View full text Add to dashboard Cite

show abstract

“…A table of a comparison of random forest accuracy for various encoding techniques shown that the one-hot encoding has a higher dimensionality than other encoding schemes [33]. Word embedding has also been considered as entity embedding in [34], this research has shown that a neural network can learn the mapping during a typical supervised training phase; therefore, with the development of entity embeddings, there has been a recent advance in categorical variable representation [35]- [38]. Moreover, when compared to the frequently used one-hot encoding, the introduction of word embeddings not only lowered memory use but also enhanced the machine learning algorithms learning ability from data [39].…”

Section: Related Workmentioning

confidence: 99%

A Deep-Learned Embedding Technique for Categorical Features Encoding

2021

View full text Add to dashboard Cite

Many machine learning algorithms and almost all deep learning architectures are incapable of processing plain texts in their raw form. This means that their input to the algorithms must be numerical in order to solve classification or regression problems. Hence, it is necessary to encode these categorical variables into numerical values using encoding techniques. Categorical features are common and often of high cardinality. One-hot encoding in such circumstances leads to very high dimensional vector representations, raising memory and computability concerns for machine learning models. This paper proposes a deep-learned embedding technique for categorical features encoding on categorical datasets. Our technique is a distributed representation for categorical features where each category is mapped to a distinct vector, and the properties of the vector are learned while training a neural network. First, we create a data vocabulary that includes only categorical data, and then we use word tokenization to make each categorical data a single word. After that, feature learning is introduced to map all of the categorical data from the vocabulary to word vectors. Three different datasets provided by the University of California Irvine (UCI) are used for training. The experimental results show that the proposed deeplearned embedding technique for categorical data provides a higher F1 score of 89% than 71% of onehot encoding, in the case of the Long short-term memory (LSTM) model. Moreover, the deep-learned embedding technique uses less memory and generates fewer features than one-hot encoding.

show abstract

“…In this regard, one hot encoding algorithm is applied for the purpose of more realistic classification of surface defects through the efficient modeling of categorical values. It is selected over the label encoding or integer encoding algorithm because in the label encoding, the higher the categorical value the more important the category, which may lead to confusion and misinterpretation of the encoded variables by the machine-learning model ( 87 , 88 ). In the one hot encoding algorithm, the categorical values are encoded as a binary vector.…”

Section: Model Developmentmentioning

confidence: 99%

Hybrid Elman Neural Network and an Invasive Weed Optimization Method for Bridge Defect Recognition

Abdelkader

Moselhi

Marzouk

et al. 2020

Transportation Research Record: Journal of the Transportation R

View full text Add to dashboard Cite

Existing bridges are aging and deteriorating, raising concerns for public safety and the preservation of these valuable assets. Furthermore, the transportation networks that manage many bridges face budgetary constraints. This state of affairs necessitates the development of a computer vision-based method to alleviate shortcomings in visual inspection-based methods. In this context, the present study proposes a three-tier method for the automated detection and recognition of bridge defects. In the first tier, singular value decomposition ([Formula: see text]) is adopted to formulate the feature vector set through mapping the most dominant spatial domain features in images. The second tier encompasses a hybridization of the Elman neural network ([Formula: see text]) and the invasive weed optimization (I[Formula: see text]) algorithm to enhance the prediction performance of the ENN. This is accomplished by designing a variable optimization mechanism that aims at searching for the optimum exploration–exploitation trade-off in the neural network. The third tier involves validation through comparisons against a set of conventional machine-learning and deep-learning models capitalizing on performance prediction and statistical significance tests. A computerized platform was programmed in C#.net to facilitate implementation by the users. It was found that the method developed outperformed other prediction models achieving overall accuracy, F-measure, Kappa coefficient, balanced accuracy, Matthews’s correlation coefficient, and area under curve of 0.955, 0.955, 0.914, 0.965, 0.937, and 0.904, respectively as per cross validation. It is expected that the method developed can improve the decision-making process in bridge management systems.

show abstract

Travel Mode Choice Prediction Using Deep Neural Networks With Entity Embeddings

Cited by 25 publications

References 44 publications

Travel mode choice prediction based on personalized recommendation model

Travel mode choice prediction based on personalized recommendation model

A Deep-Learned Embedding Technique for Categorical Features Encoding

Hybrid Elman Neural Network and an Invasive Weed Optimization Method for Bridge Defect Recognition

Contact Info

Product

Resources

About