Abstract
Machine learning has experienced notable advancements in recent times. Furthermore, this field facilitates the automation of human evaluation and processing, leading to a reduced demand for manual labor. This research paper employs data mining techniques and Knowledge Discovery in Databases (KDD) to conduct an evaluation and classification of various algorithms for pattern extraction and soil suitability prediction. The study utilizes experimental data, data transformation, and pattern extraction techniques on diverse soil samples obtained from different regions of Negros Occidental, Philippines. Specifically, the Naive Bayes, Deep Learning, Decision Tree, and Random Forest algorithms are selected for the classification and prediction of soil suitability based on the available datasets. The assessment of soil-crop suitability is based on data sourced from the Philippine Rice Research Institute, considering 14 parameters including inherent fertility, soil pH, organic matter, phosphorus, potassium, nutrient retention (CEC), base saturation, salinity hazard, water retention, drainage, permeability, stoniness, root depth, and erosion. The findings indicate that the Random Forest algorithm achieved the highest accuracy rate at 94.6% and the lowest classification error rate at 5.4%, suggesting a high level of confidence in the model's predictions. The model's predictions reveal that most soil samples in the area are only marginally suitable for banana, maize, and papaya crops. Furthermore, the study demonstrates that the majority of soil samples have a low fertility rating, which significantly impacts crop suitability. The information obtained from this study can serve as a basis for local farmers to develop improved soil management programs aimed at ensuring more productive soil. Simultaneously, it can contribute to active soil protection initiatives addressing issues such as acidity and salinity in Negros Occidental, Philippines.
Publisher
International Journal of Advanced and Applied Sciences