Traffic volume prediction using LSTM improved with rules obtained from anomalies

Download

YasinBerkGültekin_MsThesis.pdf

Date

2021-9

Author

Gültekin, Yasin Berk

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

317
views

493
downloads

Traffic volume forecasting is crucial in order to create a successful smart transportation system. The precision and timeliness of the traffic flow data are critical to the forecast's effectiveness. The lack of data has led to the usage of shallow architectures in traffic forecasting models or the creation of models based on fabricated measurement data. These models did not have a high level of success in predicting outcomes. In the world of big data, the variety and scale of the acquired traffic data has increased in lockstep with the increase in traffic density. It has a great importance for any field to be able to determine the situations that may occur in the future from the events of the past. History repeats itself constantly and in many time series problems it might be possible to make predictions for the future from what happened in the past. The LSTM structure in the Machine Learning has been implemented in order to be able to do these predictions. LSTM is a special type of recurrent neural network and its main difference from standard recurrent neural network is its success in modeling long-term information. For LSTM usage, there must be ongoing data for a certain period of time, and from this historical data the future state is predicted. In this research, we use LSTM to estimate the number of vehicles in traffic and by estimating it, we aim to see traffic congestions before they occur. In order to overcome the difficulty of LSTM when predicting vehicle count at an anomaly state, we added a new layer that puts these anomalies in a certain rule set and we used this rule set to improve accuracy of the predictions. In this study, an improved hybrid model was created by adding anomaly-based rules on top of the LSTM model, thus a model that achieves successful results even in anomaly situations.

Subject Keywords

Machine Learning, Long-Short Term Memory, Traffic Volume Prediction, Deep Learning, Anomaly Detection, Hybrid LSTM

URI

https://hdl.handle.net/11511/93167

Collections

Graduate School of Natural and Applied Sciences, Thesis

Suggestions

OpenMETU
Core

Peak-aware traffic prediction with deep learning models and a driver simulation method with probabilistic hybrid automaton Acun, Fatih; Aydın Göl, Ebru; Kalkan, Sinan; Department of Computer Engineering (2022-8) Accurately predicting traffic is crucial due to its impact on urban life in many aspects. Several statistical methods, machine learning, and deep learning approaches are applied to different traffic datasets. In general, traffic follows a stable behavior except for the morning and evening peaks which span a small period in time. To consider a prediction model to be accurate, it must demonstrate successful results during peak hours. In this thesis, a novel distance to mean weighting technique is presented th...
Traffic safety climate attitudes of road users in Germany GEHLERT, Tina; HAGEMEİSTER, Carmen; Özkan, Türker (2014-09-01) This paper validates traffic safety climate attitudes based on a representative sample of road users of all travel modes. We use the German version of the Traffic Climate Scale (TCS) which was applied in a large-scale road safety survey in 2010. A total of 1680 people were surveyed. The sample is representative for socio-demographic characteristics and travel mode choice in Germany. Factor analysis reveals a three-factor structure of traffic safety climate with the factor 'External affective demands' descri...
Detection of urban traffic patterns from Floating Car Data (FCD) ALTINTAŞI, ORUÇ; Tüydeş Yaman, Hediye; Tuncay, Kağan (Elsevier BV; 2016-09-07) Real time data collection in traffic engineering is crucial for better traffic corridor control and management. In the literature, many data collection methods have been used such as; magnetic loops, road tube counters, piezo sensors, radars, Bluetooth etc. to estimate the link occupancy, average speed or density of a corridor. More recently, Floating Car Data (FCD) has become another important traffic data source and has an increasing usage due to its lower cost and higher coverage despite its reliability ...
Traffic Network Control From Temporal Logic Specifications Coogan, Samuel; Aydın Göl, Ebru; Arcak, Murat; Belta, Calin (2016-06-01) We propose a framework for generating a signal control policy for a traffic network of signalized intersections to accomplish control objectives expressible using linear temporal logic. By applying techniques from model checking and formal methods, we obtain a correct-by-construction controller that is guaranteed to satisfy complex specifications. To apply these tools, we identify and exploit structural properties particular to traffic networks that allow for efficient computation of a finite-state abstract...
Traffic climate scale: Comparing samples from Turkey and Sweden Wallén Warner, Henriette; Öztürk, İbrahim; Özkan, Türker (2021-01-01) Traffic climate is a recent and one of the essential topics in traffic and transportation research. Various studies have examined the relations of traffic climate with driving outcomes such as accidents by using different versions of the Traffic Climate Scale (TCS). In a recent attempt, 16-items and 38-items versions of the TCS were examined in different countries. With respect to that, the present study aims to investigate the psychometric properties of 16-items and 38-items versions in samples from Turkey...

Citation Formats

Y. B. Gültekin, “Traffic volume prediction using LSTM improved with rules obtained from anomalies,” M.S. - Master of Science, Middle East Technical University, 2021.