从测试数据集中获得最佳准确性

Question

I am using Random forest classifier to classify data into 4 labels. There are a total of 20 features on which the model is being trained. I am observing an accuracy of around 45-47 % when Testing dataset is used. Although on prediction of Training dataset I'm getting an accuracy of 100%. Also I'm using the best parameters extracted using Grid Search approach. Can anyone explain why such kind of biasing between training and testing prediction. How can I enhance the scenario?

PS: I'm new to Machine Learning

Answer 1

There might be many Reasons 1)one might be, model is over fitting. you can try doing hyper parameter optimization to find the optimal value where your model performs better. 2) Since you are using accuracy as a performance parameter, You can check whether data set is balanced or not. If you are using imbalanced data set,you can use ROC,AUC Characteristics.