Implementasi Sistem IoT Monitoring Unsur Hara Tanah Berbasis Machine Learning

Abstract

Pertanian modern menuntut efisiensi pengujian kesuburan lahan. Penelitian ini merancang purwarupa Soil Integrated Tracking and Sensing atau SITRAS, sistem monitoring berbasis Internet of Things dan Machine Learning untuk rekomendasi pemupukan presisi. Sistem mengakuisisi Nitrogen, Fosfor, dan Kalium sebagai parameter utama, beserta parameter lingkungan sekunder menggunakan sensor RS485 7 in 1 yang diproses melalui mikrokontroler ESP32 S3. Guna menekan galat fisis sensor, algoritma Polynomial Ridge Regression tingkat tiga diterapkan. Model ini sukses menurunkan simpangan Mean Absolute Error sebesar 96.6 persen untuk Fosfor dan 81.9 persen untuk Kalium. Data terkalibrasi kemudian diproses algoritma K Nearest Neighbors untuk memetakan takaran pupuk Urea, SP36, dan KCl pada tanaman padi, jagung, dan kedelai. Evaluasi membuktikan prapemrosesan data adaptif ini berhasil mendongkrak akurasi ketepatan dosis secara masif. Persentase takaran SP36 dan KCl yang keduanya benar melonjak dari 10.5 persen pada sensor mentah menjadi 94.7 persen, dengan rerata akurasi benar total 97.35 persen. SITRAS terbukti menghadirkan solusi teknologi objektif guna mendukung pertanian berkelanjutan.

Modern agriculture demands efficient soil fertility testing. This research designed Soil Integrated Tracking and Sensing or SITRAS, an Internet of Things and Machine Learning based monitoring system for precise fertilization recommendations. The system acquires Nitrogen, Phosphorus, and Potassium as primary parameters, along with secondary environmental parameters using a 7 in 1 RS485 sensor processed through an ESP32 S3 microcontroller. To minimize physical sensor errors, a third degree Polynomial Ridge Regression algorithm was applied. This model successfully reduced the Mean Absolute Error deviation by 96.6 percent for Phosphorus and 81.9 percent for Potassium. The calibrated data is then processed by a K Nearest Neighbors algorithm to map Urea, SP36, and KCl fertilizer doses for rice, corn, and soybeans. Evaluation proved that this adaptive preprocessing massively boosted dose accuracy. The percentage where both SP36 and KCl doses were correct jumped from 10.5 percent on the raw sensor to 94.7 percent, with an average total correct accuracy of 97.35 percent. SITRAS provides an objective technological solution to support sustainable agriculture.