Prediction of forest unit volume based on hybrid feature selection and ensemble learning
- 10 Downloads
Aiming at the characteristics of forestry data with high dimensionality and complex samples, this paper explores an ensemble learning method suitable for predicting forest unit volume, which provides a scientific basis for forest resource management and decision-making. According to the real data provided by the National Forestry Science Data Sharing Service Platform, a FL-Stacking model based on hybrid feature selection and ensemble learning is proposed. Firstly, the model extracts features based on Filter-Lasso hybrid method, then constructs the prediction model of forest unit volume based on ensemble learning, and uses eight prediction models such as Linear SVM regression as the fusion basis model in the training set by Stacking scheme. The data are verified by 10 folds cross-validation. Finally, the fusion and optimization of the basic model are carried out. The experimental results show that the optimal accuracy of the single model is 83.81%, the multi-model predicted by FL-Stacking model is 84.55%, and the R2 value is increased by 0.74 percentage points. The comparative analysis results of different models on real data sets show that the FL-Stacking integrated prediction model proposed in this paper has a high accuracy in estimating forest unit volume, and has a great practical research value.
KeywordsPrediction of unit volume Hybrid feature selection Ensemble learning Model fusion Forest resources
This work was supported by Social Science Project of Beijing Education Commission (SM201910028017) and Capacity Building for Sci-Tech Innovation - Fundamental Scientific Research Funds of Beijing Education Commission (Grant no.19530050142). Thanks for the China National Forestry Science Data Sharing Service Platform’s Second-Class Survey and Related Data.
- 1.State Forestry Administration (2014) Results of the eighth national forest resources inventory. For Res Manage (1):1–2Google Scholar
- 6.Grossmann E (2004) Ada tree: boosting a weak classifier into a decision tree. In: Proceedings of the 2004 conference on computer vision and pattern recognition workshop. p 105Google Scholar
- 9.Heng W, Kunliang D, Xianglin T, Shuichao S, Jun CS, Pengxiang Z, Tianjian C (2015) Evaluation of site quality of natural secondary forest and artificial forest in Qinling forest region. Sci Sci Technol 51(04):78–88Google Scholar
- 13.Dong W, Zhou G, Xia L et al (1979) Quantitative theory and its application. Jilin People’s Publishing House, ChangchunGoogle Scholar
- 16.Guan BT, Gertner G (1991) Using a parallel distributed prcessing system to model individual tree mortality. For Sci 37:871–885Google Scholar
- 17.Guan BT, Gertner G (1991) Modeling red pine tree survival with an artificial neural network. For Sci 37:1429–1440Google Scholar
- 21.Yin C, Liu M, Sun F-Y et al (2016) Influencing factors of non-point source pollution of watershed based on boosted regression tree algorithm. Chin J Appl Ecol 27(3):911–19Google Scholar
- 22.Ou Q-X, Li H-K, Yang Y (2017) Factors affecting the biomass conversion and expansion factor of masson pine in Fujian Province. Acta Ecol Sin 37(17):5756–5764Google Scholar
- 23.Ou Q, Li H et al (2018) Comparison of biomass conversion and expansion factor estimation of Pinus massoniana in Fujian based on inventory data—comparison of -3 ensemble learning decision tree models. Chin J Appl Ecol 29(06):2007–2016Google Scholar
- 24.Ding L, Luo P (2017) Research on early warning of default risk of P2P online loans based on Staking integration strategy. Invest Res 36(04):41–54Google Scholar
- 25.Ye S, Wang X et al (2011) Transient stability assessment of power system based on stacking meta-learning strategy. Power Syst Prot Control 39(06):12–16Google Scholar