Figure Ilustration AI

FORMOSA NEWS - Medan - AI and Dual-Zone Sensors Revolutionize Smart Farming for Tropical Livestock Management. Researchers from Universitas Sumatera Utara and Universitas Bina Nusantara have successfully developed an artificial intelligence (AI) and Internet of Things (IoT) monitoring system to track livestock health and pen environments in real time. The 49-day study, conducted between October 4 and November 22, 2025, was led by Yudhistira Adhitya Pratama, Normalina Napitupulu, Zulhamsyah Fachrurrazi Nasution, and Adli Abdillah Nababan. Published in the Formosa Journal of Computer and Information Science (FJCIS), this research marks a significant milestone in Precision Livestock Farming (PLF). The smart sensing system provides a vital technological tool for smallholder farmers to automatically detect critical animal welfare issues like severe heat stress and toxic gas accumulation that manual inspections routinely miss.

The Hidden Costs of Traditional Farming
Goat farming underpins the rural economy across Southeast Asia. In Indonesia alone, the national herd exceeds 18 million head, mostly managed within small, traditional operations. Currently, the standard management approach relies heavily on periodic manual inspection, where a farmer walks the pen to spot visible health problems. While this manual approach works well for small herds, it fails as operations scale up or face unpredictable tropical climate shifts. Weather-driven stressors like extreme heat can quietly devastate animal productivity before physical signs appear. Furthermore, traditional stilted or slatted-floor pens concentrate animal waste directly underneath the livestock, creating invisible gradients of harmful gases like ammonia and methane that impact animal health.

Simple Tech, Deep Insights: The Dual-Zone Method
To eliminate the blind spots of manual livestock monitoring, the research team from Universitas Sumatera Utara and Universitas Bina Nusantara deployed an innovative, non-invasive dual-zone sensor network inside a slatted-floor goat pen located in North Sumatra, Indonesia. Instead of strapping expensive, stressful wearable sensors to individual goats, the researchers treated the pen itself as a single biological unit. The methodology utilized eight low-cost, interconnected IoT sensors divided across two physical zones to capture environmental and animal data simultaneously at four-second intervals:
  • The Upper Zone (Animal Platform): This area monitored the actual living space of the goats using a DHT22 temperature-humidity sensor, a KY-038 analog sound detector to capture vocalizations, and a PIR HC-SR501 motion sensor to track physical activity.
  • The Lower Zone (Waste Collection Pit): Positioned directly beneath the slatted floor, this zone featured a second DHT22 sensor alongside an MQ135 gas sensor (to detect ammonia, carbon dioxide, and nitrogen oxides) and an MQ4 sensor dedicated to tracking methane gas.
The data streams were transmitted via a NodeMCU ESP8266 microcontroller to a secure cloud database. Over the 49-day observation period, the system gathered a raw dataset. The team applied a systematic data cleaning pipeline to remove minor technical artifacts like sensor dropouts and noise, retaining 213,704 high-accuracy records (98.6% of the raw data) for AI analysis.

Machine Learning Discovers Eight Animal and Pen States
The core breakthrough achieved by Yudhistira Adhitya Pratama and his co-authors involved processing the massive environmental dataset through an unsupervised machine learning algorithm known as K-Means clustering. By evaluating the data without requiring pre-labeled training inputs, the AI automatically identified eight distinct behavioral and environmental states within the pen:
  • Severe Heat Stress (14.1% of observations): Occurring daily between 09:00 and 17:00 with a peak at 15:00, this state registered an extreme Temperature-Humidity Index (THI) mean of 92.1. The AI revealed that under this severe thermal load, the goats remained quiet to preserve energy but moved continuously around the pen to find cooler spots.
  • Moderate Heat Stress (8.6% of observations): This state also emerged during midday hours, showing a THI mean of 90.7. The livestock exhibited early-stage heat avoidance behavior characterized by moderate movement and reduced noise.
  • Post-Feeding Waste Gas Peak (3.8% of observations): Confined tightly to the evening window between 19:00 and 22:00, this state recorded the highest levels of methane gas in the pit. The AI successfully linked this peak to anaerobic fermentation happening in the fresh waste hours after the afternoon feeding.
  • Active-Vocal Daytime State (11.9% of observations): Peaking at noon, this cluster mapped normal livestock behaviors such as loud vocalizations and social interactions during feeding times.
  • Active Morning Transition (14.6% of observations): Occurring around 07:00, this state captured the animals waking up and moving actively while residual nighttime waste gases were still clearing from the pit.
  • Three Nocturnal Resting States (46.9% combined): The AI divided the nighttime resting hours into three clear baseline states differentiated purely by the gradual accumulation of waste pit gases, which peaked significantly around 03:00.
Real-World Impact and Smart Farming Recommendations
The findings from this Universitas Sumatera Utara and Universitas Bina Nusantara collaboration offer highly practical, low-cost strategies for the livestock industry, agricultural policymakers, and commercial farmers looking to optimize animal welfare and operational efficiency. Because the AI successfully separated moderate heat stress from severe heat stress, farmers can implement a graduated, energy-efficient automated response. Instead of running expensive cooling systems all day, operations can deploy moderate ventilation when the system identifies moderate heat conditions, and trigger high-power water-misting or forced cooling systems only when severe heat conditions peak between 09:00 and 17:00.

Author Profiles
Yudhistira Adhitya Pratama, S.Kom., M.Kom. is a lecturer and researcher in the Information Technology Program, Faculty of Vocational Studies at Universitas Sumatera Utara. His expertise focuses on Internet of Things (IoT) architectures and sensor data processing.
Normalina Napitupulu, S.T., M.Sc. is an academic and senior researcher in the Information Technology Program, Faculty of Vocational Studies at Universitas Sumatera Utara, specializing in embedded systems and automated environmental monitoring.
Zulhamsyah Fachrurrazi Nasution, S.T., M.I.T. is a research faculty member in the Information Technology Program, Faculty of Vocational Studies at Universitas Sumatera Utara, with expertise in cloud computing infrastructure and network management.
Adli Abdillah Nababan, S.Kom., M.T.I. is a data scientist and lecturer in the Information Systems Program at Universitas Bina Nusantara, focusing on machine learning applications and predictive analytics.

Source
Yudhistira Adhitya Pratama,Normalina Napitupulu, Zulhamsyah Fachrurrazi Nasution, Adli Abdillah Nababan (2026). IoT-Based Multi-Sensor Fusion for Goat Behavioral Pattern Recognition Using K-Means Clustering in a Smart Farming Environment. Formosa Journal of Computer and Information Science (FJCIS) 2026. Vol. 5, No. 1, Halaman 99-110
DOI: https://doi.org/10.55927/fjcis.v5i1.16599
URL: https://journal.formosapublisher.org/index.php/fjcis