Comparative Study of J48 Decision Tree Classification Algorithm, Random Tree, and Random Forest on In-Vehicle CouponRecommendation Data

Abstract

Coupons are one of the media used to increase sales and invite customers to repurchase products. A study to investigate the effectiveness of the distribution of coupons, especially coupons for restaurants and bar, can be carried out by collecting data through an in-vehicle survey. The data can then be analyzed using classification techniques in data mining. This paper presents a classification on the problem of in-vehicle coupon recommendation to determine the decision of coupon acceptance through the J48, Random Tree, and Random Forest decision tree classification algorithm. The dataset used consists of 23 attributes including the class Y attribute which indicates the receipt of coupons by customers. The performance of the three algorithms is evaluated to determine the best classification algorithm by looking at accuracy, time to build the model, and other variables that appear in the class classification experiment. The results reveal that the Random Tree classification algorithm takes the least amount of time (0.28 seconds) and has the lowest accuracy (67.38%). The J48 algorithm is more accurate than the Random Tree algorithm (72.79%) but takes significantly longer time (0.36 seconds). The Random Forest technique has the best accuracy (77.0%), but the time it takes for model creation is substantially longer than the Random Tree and J48 algorithms (10.89 seconds).