The Investigation Related to the Influence of Different Parameters Based on Random Forest for Weather Prediction

Abstract

The weather change is important in life for providing useful information that guides people's daily activities. However, in a day that machine learning is used in many fields, the weather prediction using machine learning still needs to be developed. This article uses random forest, a machine learning method that is used in classification to do the weather prediction. By changing the parameters in different values and using the evaluate metrics, this article compared the performance about random forest in different parameters. There are 4 parameters researched in this article. The first parameter is the number of trees. In this situation, the f1-score, recall and precision by the categories have the same trend with accuracy. However, the evaluate metrics about result of 1 is obviously worse than the metrics about result of 0. The second parameter is the size of data used to train. In this situation, the performance when the splitting data size is 0.1 and 0.15 is better than the other. The third parameter is the way of encoding. In this situation, using the order of the probability of rain is better than with no order or using the probability to represent the location. Fourth is to delete the unimportant features by the sort of important score. In this situation, when 4 features are deleted, the performance is worse than any others.