This paper proposes an innovative approach to trafficdensity estimation. It defines a method that focuses onreducing computational time and complexity by extractingrow, column and diagonal mean feature vectors from theimage. Then these feature vectors are used to trainclassifiers and the images are classified as low, moderate orhigh traffic situations. The system works in 2 phases:Training phase and classification phase. In the trainingphase, the image is subtracted to obtain the vehicles. Thefeatures of the subtracted image are extracted and a datasetis created. This dataset is used to train the classifier. In thesecond phase, the trained classifier is used to classify thereal-time traffic data. Finally, seven data mining classifiersare used along with total fifteen combinations of featurevectors to test the accuracy of the eighty-four variations ofthe proposed technique. The Bayes family is proved to bebetter for traffic classification. The column mean featureshave been proven better. Overall Naïve Bayes classifierwith column mean feature vector has given the betteraccuracy among experimented data mining classifiers.