An essential activity of driving is making lane changes. Depending on the driver’s motivation, lane changing events may be classified as discretionary or mandatory. Past research has shown that there is a difference in drivers’ risk-taking behavior when making discretionary and mandatory lane changes. A lane changing decision model, based on Fuzzy Inference System (FIS), has been developed with promising accuracy. This research investigates if such model can be adapted to make decisions for mandatory lane changing moves and if it is necessary to develop a new model from scratch that is dedicated to mandatory lane changes. Vehicle trajectory data of mandatory lane changing events in the NGSIM database was extracted to form a training and a test data set for comparative evaluation. First, the FIS model developed in earlier research for discretionary lane changes was directly applied to the mandatory lane changing data. Then, an Adaptive FIS (AFIS) model was implemented by adjusting a critical parameter in the FIS-based discretionary lane changing model to optimize its performance for the mandatory lane changing training data set. Additionally, new models based on Adaptive Neuro-Fuzzy Inference Systems (ANFIS) were developed for mandatory lane changes using the training data set. These models were evaluated using the mandatory lane changing test data set. Based on the test results, one of the ANFIS models is recommended, as it gave a higher overall correct decision rate compared to the existing FIS, AFIS, and other ANFIS models.

Driving at night is riskier in terms of crash involvement than it is during the day. Fortunately, it is clearly established that illumination on roadways can reduce the number and severity of nighttime crashes. However, state and municipal departments of transportation (DOTs) lack the available illumination data. Therefore, the objective of this research is threefold, as follows: (i) to develop machine learning models that use readily available roadway characteristic data to predict the severity of nighttime crashes; (ii) determine the effect that illumination has on crash severity; and (iii) develop a tool to assist DOT decision makers in collecting illumination data. To accomplish this objective, we have extracted data from the Texas Department of Transportation (TxDOT) Crash Record Information System (CRIS) database, which was then further split into a training and a test dataset. Then, seven machine learning techniques, namely binary logistic regression, k-nearest neighbors, naïve Bayes, random forest, artificial neural network, Extreme Gradient Boosting (XGBoost), and a Long Short-Term Memory (LSTM) model, were all applied to the unseen test data. The random forest model produced the most promising results by predicting severe crashes with 97.6% accuracy. In addition, we conducted a pilot study to test the collection of illumination data using a light meter. In the future, we aim to complete the development of a smartphone application, which can be used in conjunction with the random forest model presented in this paper, to collect crowdsourced illumination data and predict nighttime crash hotspots. This may assist DOT decision makers to prioritize funding for illumination at the hot spots.

Parking demand at university campuses has been an issue for decades and is gradually increasing each year. With limited capacity, space, and funds to expand parking facilities, there is a dire need to better understand parking behavior on a university campus so that universities can better utilize the limited resources available. One methodology that has been used by Metropolitan Planning Organizations to predict traveler behavior is known as travel demand modeling, where the most common modeling technique is a four-step procedure that utilizes socioeconomic data to predict current and future traffic volumes in a network (e.g., a city). This study focuses primarily on the trip generation step, and The University of Texas at Tyler campus was used as a case study. First, each parking lot on campus was assigned to a parking demand zone (PDZ) based on its proximity to classroom buildings, and the hourly arrival demand for each PDZ was measured using pneumatic tube counters. The course schedule and floor space utilization data for each building on campus were then extracted as input parameters to predict the parking demand at each PDZ. A linear regression model and two artificial neural network (ANN) frameworks were developed. One ANN model was recommended, as it had an R-squared value of 0.846. From the selected ANN model, an equation has been extracted, which has the potential to be used by course schedulers to modify the course schedule to better mitigate parking demand on campus without the need to develop new parking facilities.

When there are multiple lanes to choose from downstream of a turning movement, drivers should choose the innermost lane so that drivers at other approaches of the intersection may make concurrent turning movements in the outermost lane(s). However, human drivers do not always choose the innermost lane, which could lead to crashes with other vehicles. Therefore, predicting human driver behaviors is vital in reducing crashes, as the need to share the roadways with automated vehicles (AVs) continues to grow. In this research, various machine learning models have been used to predict the left turn destination lane choice of human-driven vehicles (HDVs) at urban intersections based on several quantifiable parameters. A total of 174 subject vehicles were extracted and analyzed in Los Angeles, California, and Atlanta, Georgia, using HDV trajectory data from the Next Generation SIMulation (NGSIM) database. Five machine learning techniques, namely binary logistic regression, k nearest neighbors, support vector machines, random forest, and adaptive neuro-fuzzy inference system, were applied to the extracted data to predict the lane choice behavior of drivers. The k nearest neighbors model showed the most promising results for the evaluated data with a correct decision score of over 93% for the unseen test data. This model may be programmed into: (i) AVs, in conjunction with sensors, to predict if an HDV is about to turn into the incorrect destination lane; and (ii) microscopic traffic simulation tools so that modelers can identify potential conflicts when HDVs do not select the appropriate destination lane.