Traffic Queue Prediction

Traffic-Queuing problems are getting more and more critical. But with advanced NLP and deep learning Algorithms we are much better in traffic safety. But more and more critical measures are being taken. This is one of those projects in which I and one of my classmate, a PHD candidate focused on classifying traffic situation at a traffic signal with minimal 24 hours of telematic data from MioVision API extraction. My classmate work for a City of Detroit and they are doing pretty creative work with optimizing traffic conditions as Detroit. Kudos to my friend for working with me at this university project.

We worked step by step like any data Analytics/Data Science Project flows. We started by downloading sample data from MioVison API in JSON format. Converted it locally into CSV. We then employed some EDA with python to visualize correlation plots, bar graphs, scaling and normalizing the continuous data. Then we applied some machine learning classification algorithms like Logistic regression, MLP, SVM, KNN.

Battle of Algorithms

We predicted the Traffic state as NQ: No Queue, LQ: Low Queue and FQ: Full Queue and based on the confusion matrices and precision-recall matrices, we reached a conclusion that KNN was a suitable prediction.

The K War

After choosing KNN to predict classes we still needed more analysis on the optimum k-value choice. So we plotted the Number of neighbors (K) against the training and test accuracy to reach conclusion on K= 5

After this battle of algorithms we also plotted the ROC curve and classification boundary for later insights.

You can read the detail project below: