Effects of agricultural land-use types on soil physicochemical parameters in Central Vietnam using machine learning

Exploring differences in soil parameters across land-use types is important for sustainable land management and agricultural development, particularly in Central Vietnam, where soil degradation is worsening owing to intensive farming, erosion, and unsustainable practices. This study examined principal component analysis (PCA) and Random Forest (RF) in selecting a minimum data set (MDS) for soil quality index (SQI) development. We then applied analysis of variance (ANOVA) to evaluate the differences in soil quality among agricultural land-use types (ALUTs) (i.e., rice, cassava, and rubber). Eleven soil properties (organic matter, total nitrogen, total phosphorus, total potassium, pH, sand, silt, clay, cation exchange capacity, salinity, and sulfate) were measured from 94 samples. The results showed higher stability for the RF-based MDS (stability index (SI) = 0.939) than for the PCA-based MDS (SI = 1.186), given that smaller SI values indicate greater stability. ANOVA results revealed statistically significant differences in SQI RF (F RF value = 5.829, p RF = 0.004) and SQI PCA (F PCA value = 6.273, p PCA = 0.002) among the ALUTs. The Duncan test showed that rice fields had the highest SQI, which was significantly greater than that of cassava and rubber in the PCA and RF methods. In conclusion, ALUTs considerably affect soil quality. The RF approaches showed better performance than PCA. We recommend using RF as a reliable and effective tool for soil quality assessment in different land-use types in Central Vietnam and other areas. • A machine learning framework was explored for minimum data set (MDS) selection. • A comparison between principal component analysis (PCA) and Random Forest (RF) was conducted. • The RF-based MDS showed higher stability than the PCA-based MDS. • Rice soils exhibited higher SQI compared to cassava and rubber soils.