Open Access

ARTICLE

Application of the random forest algorithm for predicting the persistence of seed banks in the Horqin Sandy Land, China

by Tang Y, SS Jin

1 School of Life Science, Liaoning University, Shenyang, 110036, China.
2 School of Management, Fudan University, Shanghai 200433, China.

Address correspondence to: Shusong Jin, e-mail:

Phyton-International Journal of Experimental Botany 2018, 87(all), 280-285. https://doi.org/10.32604/phyton.2018.87.280

Abstract

Persistent seed banks have been detected in the Horqin Sandy Land, China using experimental methods. In this study, we used seed traits (i.e. seed mass and seed shape) to predict the persistence of seed banks using the random forest algorithm. The results showed that the mean decrease in accuracy for seed mass and seed shape was 18.26 and 9.90, respectively, suggesting that seed mass was a better predictor than seed shape. With increasing seed mass, the log of P (where P is the ratio of the number of votes selecting existence of a persistent seed bank to the number of votes selecting absence of a persistent seed bank) gradually decreased. It also changed from positive to negative when the seed mass reached 10 mg. This indicates that small-seeded species tend to have a persistent seed bank, and larger-seeded species tend to germinate in the current year. Furthermore, a seed mass of 10 mg was the dividing point for predicting the persistence of seed banks in this region. The log of P decreased as the seed shape decreased in the range from 0.07 to 0.17, indicating that seed shape (0.07–0.17) is negatively related to the existence of a persistent seed bank. It is suggested that the random forest algorithm is a useful tool for predicting the persistence of seed banks.

---

Keywords

---

Citations