EMLARDE tree: ensemble machine learning based random de-correlated extra decision tree for the forest cover type prediction

Forest cover type prediction is used for the forest management organizations. It also get the insight on area of the forest cover up to date and development lack in present time. Classification of the forest area and type of trees could eventually help in maintaining the eco system and to get inference on deforestation. In present scenario this problem gains more attention hence to retain the climate change impact forest cover type and area prediction would help a lot. This paper proposes a novel ensemble machine learning based random de-correlated extra decision tree model for the forest cover type prediction. The tree based classifiers perform well in prediction of the forest cover data. Many researchers use tree based classifiers for the problem. Even though the enhancement of the accuracy seems to be lower in the multi-class classification problem. So, this research proposes the Extra random de-correlated decision tree method for the prediction of the forest cover. The results the multiple de-correlated decision trees are aggregated for the final classification. This proposed method is the ensemble based method. In ensemble machine learning method combines several base optimal results in order to produce one final optimal result. A decision tree follows a simple predictive outcomes based on the series of the cause and effect values. While adopting the decision tree models the user has to follow the factors including the variable on which the decision to be taken and threshold for deciding the class. Instead of depending on one tree for decision making, multiple tree split criteria can be considered. Also these ensemble based machine learning allow to fine tune the predictor variable based on the feature to use and split criteria. The random forest based methods follows the bagging strategy. It has a major role in the split aspect and decision-making aspect in significant manner. This machine learning model decides where to split based on random selection of features. Random forest tree methods have a uniqueness where each split can be done through scrutiny of different features. This paper proposes the ensemble machine learning based random de-correlated extra decision tree model for the forest cover type prediction. This algorithm especially suits the problem for the multiclass classification nature. Forest cover type prediction helps in identifying the wilderness type and total area of the forest predicted and available. The dataset considered for the paper is from the UCI Machine Learning repository. It contains various features including elevation, slop, aspect, vertical and horizontal distance to hydrology, fire points and roadways, hill shade, wilderness area, soil type and cover type. Initially the preprocessing is done in the data set by identifying the missing values, outlier detection and formatting data. Later the exploratory analysis is carried out using the Pearson correlation coefficients aspect. Then three machine learning techniques: Multiclass SVM, Boosting and proposed EMLARDE were deployed. The accuracy of the proposed EMLARDE method outperforms the other two algorithms. The proposed algorithm performs well for this multiclass classification.