The national fertiliser policies in South Korea aim to provide guidance to farmers for efficient fertiliser application and thus rely on the annual collection and analysis of soil samples. Providing timely soil analysis results remains a challenge, as wet laboratory analysis is time-consuming and expensive. This study represents a pioneering effort in South Korea, by investigating mid-infrared (MIR) spectroscopy for accurate soil properties prediction and its application in developing fertiliser recommendations for several crop types. Additionally, we examined the time efficiency of MIR spectroscopy compared to conventional analytical methods. A total of 567 soil samples from diverse soil and land use types (paddy, upland, orchard, and greenhouse fields) in South Korea (0–20 cm depth) were collected and scanned using an MIR spectrometer. Four machine learning algorithms (partial least squares regression, support vector machine, cubist, and random forest) were trialled and compared for their prediction accuracies using 15-fold cross-validation for eight essential soil properties: organic matter, total nitrogen (N), available phosphorus (P), pH, exchangeable calcium (Ca), potassium (K), magnesium (Mg), and available silica. Results demonstrated robust predictive performance (R2 > 0.70) across the selected soil properties, with organic matter and total nitrogen exhibiting excellent accuracy (R2 > 0.9). Compared with conventional analysis, the average difference in fertiliser application recommendation for seven crops using MIR prediction was 3.8 % for N, 13.9 % for P and 8.1 % for K. Based on the measurement of 11 soil properties, analysis using MIR spectroscopy was about 12 times faster than conventional methods. The study demonstrates the potential of this approach to revolutionise soil analysis protocols, offering a more efficient and cost-effective solution for sustainable agricultural practices in South Korea.