An Improved Algorithm based on KNN and Random Forest
Jun Liang, Qin Liu, Nuihua Nie, Biqing Zeng, Zan‐Bo Zhang · Proceedings of the 3rd International Conference on Computer Science and Application Engineering · 2019
This paper gives an improved algorithm called RFDKNN based on an enhanced KNN (K-Nearest Neighbor) and random forest. First, RFDKNN sorts features based on importance through Gini index and a random forest algorithm. Then it deletes some unimportant features based on this sort in a certain proportion r. Finally, it uses an enhanced KNN algorithm dynamically selecting the optimal nearest neighbor number and distance function to make the distance between two samples closer to true value. Experiments are carried out on the 20 data sets from UCI Machine Learning repository. The results show that compared with other r values, RFDKNN of r=0.7 can obtain a relatively satisfactory classification accuracy. Compared with Naive Bayes, Adaboost, Random Forest, RRSB, W-KNN, dwh-FNN and LI-KNN, RFDKNN has higher classification accuracy on most data sets, especially on large data sets such as Pendigits and Letter.