Enhancing LSTM Prediction of Vehicle Traffic Flow Data via Outlier Correlations

Fitters, Wesley; Cuzzocrea, Alfredo; Hassani, Marwan

doi:10.1109/compsac51774.2021.00039

Cited by 14 publications

(4 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The results were as follows: the recovery rate oscillated, the correct detection rate increased, and the accuracy decreased when the number of record increased. Fitters et al [40] proposed an outlier-enriched long short-term memory framework to detect contextual outliers. Data were collected from the urban traffic network of Hague, and the results were compared to those resulting from naïve multivariate long short-term memory in terms of running time.…”

Section: Deep Learningmentioning

confidence: 99%

“…Thus, duplicated data could be considered as a collective outlier, which would increase the false-positive rate. Furthermore, both Fitters et al [40] and Albattah et al [45] detected contextual outliers without tackling missing data. In addition, Liu et al [29] and Kulanuwat et al [54] both dealt with outliers before dealing with the missing value.…”

Section: Rq2: Which Data Cleaning Issue Is Most Commonly Discussed Du...mentioning

confidence: 99%

See 1 more Smart Citation

Cleaning Big Data Streams: A Systematic Literature Review

2023

View full text Add to dashboard Cite

In today’s big data era, cleaning big data streams has become a challenging task because of the different formats of big data and the massive amount of big data which is being generated. Many studies have proposed different techniques to overcome these challenges, such as cleaning big data in real time. This systematic literature review presents recently developed techniques that have been used for the cleaning process and for each data cleaning issue. Following the PRISMA framework, four databases are searched, namely IEEE Xplore, ACM Library, Scopus, and Science Direct, to select relevant studies. After selecting the relevant studies, we identify the techniques that have been utilized to clean big data streams and the evaluation methods that have been used to examine their efficiency. Also, we define the cleaning issues that may appear during the cleaning process, namely missing values, duplicated data, outliers, and irrelevant data. Based on our study, the future directions of cleaning big data streams are identified.

show abstract

Section: Deep Learningmentioning

confidence: 99%

Section: Rq2: Which Data Cleaning Issue Is Most Commonly Discussed Du...mentioning

confidence: 99%

Cleaning Big Data Streams: A Systematic Literature Review

2023

View full text Add to dashboard Cite

show abstract

“…In recent years, the research on optimizing trajectory data quality mainly focuses on noise (outlier) recognition and vehicle trajectory reconstruction. Fitters et al [19] propose a novel framework for traffic flow prediction, OE-LSTM (Outlier-Enriched LSTM), to identify outliers that deviate from regular traffic flow. Fard et al [20] proposed a two-step technique based on wavelet analysis, primarily by using a wavelet transform with unique processing methods to identify and modify outliers.…”

Section: Introductionmentioning

confidence: 99%

Two step vehicle trajectory reconstruction strategy based on an improved outlier detection and data smoothing algorithm

Cao

2024

International Conference on Smart Transportation and City Engineering (STCE 2023)

View full text Add to dashboard Cite

Vehicle trajectory data contains abundant traffic information and plays a key role in traffic flow research. Due to unsuitable sensor position or perspective, there are some local errors in existing vehicle trajectory data sets. To solve this problem, a two-step vehicle trajectory data reconstruction strategy is proposed in this paper. In the first step, an outlier detection algorithm combines the cluster-based local outlier factor (CBLOF) and the k-means clustering algorithm (K-CBLOF) is proposed to identify outlier in object dataset. In the second step, Savitzky-Golay (SG) filter is selected to smooth the data flow containing noise. In this paper, the prototype data of I-80 highway in the open source trajectory data set NGSIM is used for case study. Taking vehicle 1 as an example, the speed and acceleration data of vehicle 1 are reconstructed. Time Windows with length of 11 and 21 are selected respectively to implement SG filter. The results show that when the time window value is 21, the trajectory data shows a better smoothed result, and the data noise can be obviously eliminated, which verifies that the trajectory data smoothing strategy proposed in this paper has a certain reliability.

show abstract

“…Among them, LSTM can capture time series characteristics over a longer period, and overcome the problem of gradient disappearance. Hence, this model is considered effective for long-term time series predictions [15][16][17][18][19], such as pedestrian trajectory predictions [20], and traffic flow predictions [21]. The RNN circulating neural network has a memory function, i.e., calculating the state at the current time point depends on the calculation results at the previous time point.…”

Section: Introductionmentioning

confidence: 99%

Novel Neural-Network-Based Fuel Consumption Prediction Models Considering Vehicular Jerk

Zhang,

Ya,

et al. 2023

Electronics

View full text Add to dashboard Cite

Conventional fuel consumption prediction (FCP) models using neural networks usually adopt driving parameters, such as speed and acceleration, as the training input, leading to a low prediction accuracy and a poor correlation between fuel consumption and driving behavior. To address this issue, the present study introduced jerk (an acceleration derivative) as an important variable in the training input of four selected neural networks: long short-term memory (LSTM), recurrent neural network (RNN), nonlinear auto-regressive model with exogenous inputs (NARX), and generalized regression neural network (GRNN). Furthermore, the root-mean-square error (RMSE), relative error (RE), and coefficient of determination (R2) were used to evaluate the prediction performance of each model. The results from the comparison experiment show that the LSTM model outperforms the other three models. Specifically, the four selected neural network models exhibited an improved accuracy in fuel consumption prediction after the jerk was added as a new variable to the training input. LSTM exhibited the greatest improvement under the high-speed expressway scenario, in which the RMSE decreased by 14.3%, the RE decreased by 28.3%, and the R2 increased by 9.7%.

show abstract

Enhancing LSTM Prediction of Vehicle Traffic Flow Data via Outlier Correlations

Cited by 14 publications

References 31 publications

Cleaning Big Data Streams: A Systematic Literature Review

Cleaning Big Data Streams: A Systematic Literature Review

Two step vehicle trajectory reconstruction strategy based on an improved outlier detection and data smoothing algorithm

Novel Neural-Network-Based Fuel Consumption Prediction Models Considering Vehicular Jerk

Contact Info

Product

Resources

About