A Self-Configurable IoT-Based Monitoring Approach for Environmental Variables in Rotational Grazing Systems

Abstract

Shaded resting zones in rotational grazing systems are prone to thermal stress due to limited ventilation and the congregation of animals during peak heat periods. Addressing these challenges requires sensing solutions that are not only accurate but also capable of adapting to dynamic environmental conditions and energy constraints. In this context, we present the development and simulation-based validation of a self-configurable IoT protocol for adaptive environmental monitoring. The approach integrates embedded machine learning, specifically a Random Forest classifier, to detect critical conditions using synthetic data of temperature, humidity, and CO₂. The model achieved an accuracy of 98%, with a precision of 98%, recall of 85%, and F1-score of 91% in identifying critical states. These results demonstrate the feasibility of embedding adaptive intelligence into IoT-based monitoring solutions. The protocol is conceived as a foundation for integration into physical devices and subsequent evaluation in farm environments such as rotational grazing systems.

1. Introduction

In beef cattle farming systems, particularly those based on rotational grazing, shaded resting areas play a critical role in mitigating thermal stress and maintaining animal welfare [1]. However, these shaded resting areas are highly sensitive to variations in key environmental parameters—especially temperature and relative humidity [2]. Rapid increases in ambient temperature, combined with low humidity levels, can lead to heat stress conditions that negatively impact cattle health, feed intake, and productivity [3].

Rotational grazing is a pasture management strategy in which cattle are periodically moved between subdivided paddocks to optimize forage regrowth, soil recovery, and animal welfare. Unlike continuous grazing, this system allows pastures to rest after being grazed, reducing overgrazing and promoting more sustainable use of forage resources [4]. Key guidelines include paddock division, resting periods, and adjusting stocking rates according to forage availability. The scale of implementation varies considerably: small herds (10–50 cattle) are often managed with fewer paddocks and shorter cycles, while larger herds (hundreds of cattle) require more extensive infrastructure and monitoring. Although our study does not target a specific herd size, we emphasize that the proposed Internet of Things (IoT) monitoring system is scalable and adaptable to different operational contexts.

Conventional environmental monitoring strategies often fail to react to these dynamic changes, either due to fixed data acquisition rates or lack of contextual awareness [5]. This limitation is especially critical in extensive grazing systems, where the continuous deployment of human or high-energy resources is impractical.

Recent reviews highlight that IoT-based surveillance systems are increasingly applied in livestock farming, integrating machine learning to support animal health and welfare [6,7]. These studies emphasize the potential of digital tools and intelligent sensors to transform livestock management. However, most reported systems focus primarily on general health monitoring or environmental sensing without incorporating adaptive mechanisms at the edge level. In particular, self-configuration strategies—where sensor nodes autonomously adjust sampling frequency, communication protocols, and energy usage in response to environmental stressors—remain largely unexplored in practical deployments.

This research proposes a novel approach using self-configurable IoT devices capable of monitoring temperature and humidity in real time. These devices employ machine learning models to assess environmental risk based on these two variables, enabling the system to autonomously adapt its sampling frequency and communication protocol. For instance, if the temperature exceeds 28 °C and humidity drops below 60%, the sensor node increases its sampling rate and switches to a long-range communication protocol (e.g., LoRaWAN) to ensure continuous data delivery from critical areas. The adaptive logic also considers battery status to maintain operational efficiency, reducing frequency when conditions are stable or energy is limited. This self-configuration mechanism allows the sensors to respond proportionally to environmental threats, enhancing both animal welfare and system autonomy.

This abstraction highlights that the novelty of our work lies in the self-configurable property of the sensing nodes, rather than in a particular hardware implementation. By decoupling the adaptation logic from the physical platform, the system ensures portability, reproducibility, and future extensibility in precision livestock farming applications.

By focusing on temperature and humidity as primary indicators, this work contributes to the development of intelligent sensing infrastructure for precision livestock farming (PLF), offering an effective and energy-efficient solution to environmental risk detection in shaded rotational grazing zones.

The remainder of this paper is structured as follows: Section 2 reviews related work on IoT-based environmental monitoring and adaptive sensing in precision livestock farming. Section 3 outlines the theoretical background, including temperature–humidity stress indicators and self-configuration mechanisms. Section 4 describes our proposed system architecture, machine learning model, and autoconfiguration logic. Section 5 presents the experimental design, simulation setup, and evaluation metrics. Finally, Section 6 summarizes the main findings and proposes future research directions to enhance intelligent livestock monitoring systems.

2. Related Work

Environmental monitoring in livestock farming has gained attention in recent years due to its direct impact on animal welfare and production efficiency.

Santolini et al. [8] implemented an air monitoring system in a dairy farm focused on tracking gaseous emissions to improve environmental conditions and animal welfare. Their Smart Monitoring System enables the collection of environmental data both indoors and outdoors. While comprehensive, their system does not implement self-configuration or embedded machine learning models on edge devices, unlike our approach.

Li et al. [9] reviewed the relationship between environment and animal welfare, highlighting the importance of environmental stress management. Their work emphasizes the need for adaptable environments but does not propose a specific implementation or intelligent sensing framework.

Pereira et al. [10] presented an IoT-based monitoring system for poultry farms, validating its performance against commercial equipment. The focus was on affordability and sensor accuracy rather than adaptive or intelligent configuration in response to environmental changes.

Mohan et al. [11] developed “Animo”, an IoT system for monitoring livestock health in real time. Their system uses multiple sensors and validates its ability to detect health trends. However, their system does not feature embedded machine learning or autonomous reconfiguration mechanisms.

Pillewan et al. [12] designed an IoT system for monitoring environmental parameters in domestic animal shelters, focusing on alerting farmers to dangerous conditions and unauthorized movement. Their system enhances security and monitoring but lacks adaptive sensing and intelligent sampling control.

Unold et al. [13] proposed a multisensory system for non-invasive welfare monitoring of laboratory animals using low-cost microcontrollers. Their system integrates various environmental sensors and a camera, focusing on compliance with laboratory regulations. Although similar in intent, their setup is static and lacks adaptive sensor behavior in the field.

In addition, IoT monitoring systems have been proposed for dairy cattle by Tangorra et al. [14] and for environmental sensing in multi-species farms by Provolo et al. [15]. Similarly, robust, long-range LoRa-based systems have been tested in alpine pastures and remote locations by Schulthess et al. [16], demonstrating the potential of low-power wide-area networking technologies for livestock monitoring. While these approaches provide valuable insights into precision livestock farming, they generally lack self-configuration mechanisms or embedded edge intelligence capable of adapting sensor behavior in real time to environmental changes.

A comparative analysis of the reviewed studies shows that most existing IoT-based systems focus on either reliable data acquisition [8,10], welfare monitoring in static environments [12], or cloud-dependent health tracking [11,13]. Even in more recent deployments leveraging long-range communication and multi-sensor integration [14,15,16], adaptive mechanisms for self-configuration at the edge are not addressed.

In contrast to these studies, our system introduces a novel self-configurable architecture that integrates a Random Forest classifier directly into the sensor node. Unlike prior approaches that rely on fixed acquisition rates or cloud-based analysis, our system embeds machine learning models at the edge to dynamically adapt sensing frequency and communication protocols according to real-time environmental conditions and energy constraints. This edge-level intelligence enables immediate responsiveness to thermal stress events in shaded resting areas, while also optimizing energy efficiency under stable conditions.

3. Theoretical Framework

3.1. Precision Livestock Farming

PLF refers to the application of advanced technologies, such as sensors, machine learning, and real-time data analysis, to monitor and manage livestock production systems with greater accuracy and responsiveness. The objective of PLF is to improve animal welfare, enhance productivity, and optimize resource utilization by enabling informed, data-driven decisions at the individual or group level [17].

In the context of beef cattle farming, PLF systems facilitate continuous monitoring of key parameters such as animal movement, feeding behavior, body condition, and environmental conditions (e.g., temperature, humidity, and air quality) [18]. These systems are particularly valuable in rotational grazing schemes, where animals move frequently between paddocks and are exposed to varying microclimates [19].

By incorporating IoT devices and automated decision-making mechanisms, PLF allows farmers to respond proactively to risks, such as heat stress or poor ventilation in shaded areas [20,21]. This enhances not only animal health and welfare but also the sustainability and efficiency of livestock operations [22,23].

3.2. Properties of Self-* Systems

Self-* systems, also known as autonomic systems, are designed to operate with a high degree of independence by dynamically adjusting their behavior in response to internal or external stimuli [24]. These systems reduce the need for human intervention, making them particularly suitable for environments where manual management is costly or impractical, such as remote agricultural fields or distributed livestock production systems [25].

As defined by Kephart and Chess [26], autonomic systems exhibit the following core properties:

• Self-configuration:The ability to automatically configure and reconfigure components based on the current context or operational state. In IoT applications, this might include adjusting the sampling frequency, modifying communication protocols, or enabling energy-saving modes depending on sensor readings and energy levels.
• Self-optimization: The ability to monitor performance metrics and proactively make adjustments to improve operational efficiency. In the case of IoT devices in livestock systems, this may involve optimizing energy consumption while maintaining reliable data acquisition.
• Self-healing: The capability to detect, diagnose, and recover from faults automatically. Although not a primary focus of this study, self-healing is essential for long-term autonomous deployments in rural and hard-to-access environments.
• Self-protection: The ability to detect and defend against malicious attacks or unexpected disturbances that may compromise the system’s functionality or data integrity.
• Self-awareness: The ability of the system to be aware of its internal state and surrounding environment, enabling context-aware decision-making. This property supports all other self-* capabilities by serving as the foundation for adaptive behavior.

3.3. Environmental Stress Indicators in Cattle

Environmental stress, particularly thermal stress, represents a major challenge in beef cattle production systems [27]. Cattle are homeothermic animals that rely on physiological and behavioral mechanisms to maintain a stable internal temperature [28]. However, their ability to dissipate excess body heat is limited, especially under conditions of elevated ambient temperature and humidity [29].

Heat stress occurs when the environmental conditions exceed the animal’s capacity to maintain thermal equilibrium [30]. This leads to increased respiration rate, reduced feed intake, altered metabolic functions, and compromised immune response [31]. In the context of livestock operations, these physiological impacts often translate into decreased productivity, lower weight gain, fertility issues, and, in extreme cases, mortality [32].

Resting areas such as shaded zones or holding pens are particularly vulnerable to heat accumulation, especially in tropical or subtropical climates [33]. Poor ventilation and crowding can exacerbate the buildup of heat and humidity, creating suboptimal environments where cattle are exposed to chronic stress conditions [34].

4. The Proposed Approach

4.1. Hardware-Agnostic Self-Configuration Design

A key aspect of our proposal is that the system design is hardware-agnostic. Instead of binding the solution to a specific microcontroller, radio, or sensor model, we describe the self-configuration capability in terms of abstract interfaces and operational profiles. This ensures that the approach can be implemented across heterogeneous IoT platforms while preserving its adaptive behavior.

• Abstract Interfaces: Each node exposes a minimal set of functions—read() for data acquisition, predict() for on-node inference, and tx(mode) for data transmission. These interfaces allow the adaptation logic to remain independent of the underlying hardware.
• Operational Profiles: Three generic profiles govern system behavior: (i) Normal for stable conditions, (ii) Critical for detected risk situations, and  (iii) Energy-Saving when battery levels are low. Each profile specifies sampling rate, communication mode, and transmission priority, but does not prescribe a specific technology.
• Adaptation Logic: The Random Forest classifier provides a binary state (normal/critical). Combined with battery status, the adaptation engine transitions between operational profiles. Transitions are governed by hysteresis rules to avoid oscillations.
• Deployment Flexibility: Because the system is defined through abstract layers and not through specific hardware, it can be deployed with different sensing modules (e.g., various temperature or CO₂ sensors), communication technologies (WiFi, LoRa, Zigbee, etc.), and energy sources (battery, solar), without altering the self-configuration mechanism.

4.2. System Architecture

The proposed architecture consists of three functional layers—Sensing, Autonomous Configuration, and Data Management—that collaborate to ensure real-time, adaptive environmental monitoring in cattle rotational grazing zones (see Figure 1).

• Sensing Layer: This layer includes multiple sensor nodes deployed in the field. Each node measures key environmental variables such as temperature, humidity, and CO₂ concentration. The sensors operate autonomously and generate time-stamped observations. These nodes are deployed in strategic zones, such as shaded resting areas or holding corrals, where animals rest and environmental conditions are prone to stress events.
• Autonomous Configuration Layer: This layer integrates a machine learning-based anomaly detection component, specifically a Random Forest classifier trained to predict critical conditions from sensor inputs. Each sensor node embeds this model and applies it in real time to incoming data. Based on the prediction (critical vs. normal), an adaptation engine modifies the sensor’s configuration:

  -

  In critical conditions (e.g., high temperature and low humidity), the sampling frequency is increased (e.g., from 10 s to 2 s), and a low-latency communication protocol such as LoRaWAN is selected.

  -

  In low-risk or low-battery scenarios, the sensor switches to energy-saving modes (e.g., longer sampling intervals or reduced bandwidth protocols like LoRa).

  For implementation purposes, a binary variable named critical_zone is defined, which indicates whether the system is operating under critical environmental conditions (1) or normal conditions (0). This variable is used internally to trigger adaptive configuration decisions at the sensor node. This adaptive behavior reduces energy consumption and extends device lifespan while ensuring that dangerous conditions are captured with high temporal resolution.
• Data Management Layer: Once data is captured and classified, it is transmitted to an edge gateway, which acts as a local processing hub. This layer handles aggregation, storage, and visualization of incoming data. The gateway may forward data to a remote dashboard or cloud-based analytics platform, where long-term trends can be monitored and alarms issued when sustained environmental stress is detected.

Figure 1 illustrates the complete flow from raw data acquisition to configuration adaptation and remote management. This modular design supports deployment in isolated areas with limited connectivity and power resources, making it well-suited for real-world cattle production environments.

4.3. Synthetic Data Generation

To rigorously evaluate the proposed system under diverse and realistic scenarios, we generated a synthetic dataset that emulates environmental dynamics in shaded rotational grazing zones. Unlike simple random sampling, our procedure integrates deterministic cycles, stochastic noise, and stress-inducing events to approximate real-world variability while preserving reproducibility.

4.3.1. Dataset Composition

The dataset comprises over 12,000 records collected from four virtual sensor nodes (sensor_id = 1–4). Each record contains the following fields: timestamp, temperature, humidity, CO₂, pressure, critical_zone (binary label), and type_event (categorical variable indicating normal or stress conditions).

4.3.2. Environmental Patterns

• Temperature: Modeled with a sinusoidal diurnal cycle peaking at midday, plus Gaussian noise ($\mathsf{\mu }=0$, $\sigma =1$) to emulate sensor inaccuracies.
• Humidity: Initialized at 65% and modeled with a slow downward daily trend and stochastic variations ($\sigma =3$).
• CO₂: Centered at 420 ppm with random noise ($\sigma =10$) and occasional spikes to mimic accumulation due to crowding.
• Pressure: Simulated around 1012 hPa ($\sigma =4$), representing ambient atmospheric conditions.

4.3.3. Critical Events

To reproduce stress-inducing microclimates, abrupt changes were injected in approximately 15% of the samples, following empirical thresholds reported in cattle heat stress literature:

• Temperature exceeding 34 °C;
• Humidity falling below 50%;
• CO₂ concentration rising above 500 ppm.

Whenever one of these conditions was met, the binary label critical_zone was set to 1. In addition, the categorical field type_event was updated from normal to thermal_stress.

4.3.4. Noise and Missingness

To emulate common IoT constraints and ensure robustness testing:

• 5% of values were set as missing at random to mimic packet loss or sensor downtime.
• 10% of labels were flipped to simulate annotation uncertainty, reflecting imperfect ground truth.
• Timestamps were jittered by ±5 s to reproduce asynchronous sensor clocks.

All simulation scripts were implemented in Python 3.10 and parameterized by random seeds for full reproducibility. The dataset and source code are openly available at https://github.com/devraxielh/Self-Configurable-IoT-Devices-PLF-Environmental-Monitoring (accessed on 12 September 2025), enabling replication and benchmarking in future PLF research.

This synthetic data generation strategy strikes a balance between controlled experimentation and ecological validity. By explicitly modeling diurnal cycles, random perturbations, and critical stress events, it allows for a rigorous evaluation of the self-configuration mechanisms under conditions that approximate real-world shaded grazing environments. The dataset thus provides a reliable foundation for assessing adaptive sensing in Precision Livestock Farming.

4.4. Model Training and Adaptive Feedback Loop

This subsection integrates the offline training pipeline of the predictive model and the underlying assumptions that guided its design and evaluation.

4.4.1. Model Training Procedure

Table 1. Offline model training for environmental zone classification.

Step	Description
1	Remove incomplete rows with missing values from D
2	Normalize or standardize features in $x_i$
3	Split dataset D into $D_{train}$ (80%) and $D_{test}$ (20%)
4	Train Random Forest classifier f using $D_{train}$
5	Evaluate f on $D_{test}$ using accuracy, precision, recall, and F1-score
6	Save the trained model f for deployment on sensor nodes

describes the offline training procedure used to generate the predictive model that enables zone classification. This model is embedded in the sensor node and serves as the basis for self-configuration decisions.

4.4.2. Adaptive Feedback Loop

The adaptive feedback loop governs the behavior of each sensor node deployed in the grazing environment. Algorithm 1 outlines the decision-making cycle executed by the sensor in real time.

Algorithm 1 Adaptive Feedback Loop in Self-Configurable Sensor

1: while sensor is active do 2:     Read measurements $x_t\leftarrow \{\mathrm{temperature},\mathrm{humidity},{\mathrm{CO}}_2\}$ 3:     Predict label ${\widehat{y}}_t\leftarrow f\left(x_t\right)$             ▹ Random Forest classifier 4:     if ${\widehat{y}}_t$ is critical then 5:         Set frequency ← 2 seconds 6:         Set protocol ← LoRaWAN 7:     else if battery level $<30\%$ then 8:         Set frequency ← 30 seconds 9:         Set protocol ← LoRa 10:     else 11:         Set frequency ← 10 seconds 12:         Set protocol ← WiFi 13:     end if 14:     Transmit data to edge gateway 15:     Update battery level 16:     Wait for next cycle based on current frequency 17: end while

The algorithm begins by reading environmental data from the sensor, which includes temperature, humidity, and CO₂ concentration. This data is passed through a trained classifier f (e.g., Random Forest) to assess whether the environmental condition is critical (e.g., heat stress).

If a critical condition is detected, the sensor increases its sampling frequency to 2 s and switches to a low-latency protocol (LoRaWAN) to ensure rapid data transmission. If the battery level is low (under 30%), it compensates by reducing frequency and switching to a low-energy protocol (LoRa). Under normal conditions, a standard configuration is applied.

This cycle continues indefinitely, allowing the sensor to dynamically adjust its behavior based on both external stimuli and internal constraints (such as battery).

The feedback loop allows the system to maintain a balance between responsiveness and energy efficiency, which is crucial for remote deployments in cattle grazing areas where maintenance is infrequent and power resources are limited.

4.5. Study Assumptions and Limitations

This section outlines the main assumptions guiding the study and acknowledges its limitations, providing transparency on the scope of the proposed system and directions for future validation.

4.5.1. Assumptions for the Study

This study assumes that the shaded resting zones in rotational grazing systems represent critical environment where cattle tend to concentrate during peak heat hours. The synthetic dataset was designed to reproduce realistic fluctuations in temperature, humidity, and CO₂ based on diurnal patterns and stress-inducing conditions. We also assume that herd size and pasture area may vary, but the adaptive IoT framework is scalable and can be applied to both small and large herds with proper adjustment of paddock division and sensor distribution. Moreover, the system presumes access to low-cost IoT hardware and basic connectivity infrastructure, which are increasingly available in precision livestock farming contexts.

4.5.2. Limitations

The primary limitation of this work is that the validation was conducted using synthetic data rather than real-world measurements. While the simulation incorporated variability, noise, and missing values to approximate field conditions, actual deployments may introduce additional complexities such as hardware malfunctions, communication losses, and unpredictable animal behavior. Another limitation is that the current model only considers a restricted set of environmental variables (temperature, humidity, and CO₂); integrating additional parameters such as wind speed, solar radiation, or physiological indicators could enhance detection of critical conditions. Finally, large-scale field validation with different herd sizes and grazing systems will be necessary to confirm the generalizability and robustness of the proposed self-configurable IoT system.

5. Experiments and Results

5.1. Experimental Setup

The experimental setup was designed to simulate sensor deployment in a shaded rest area of a cattle rotational grazing system. The goal was to generate controlled environmental data under realistic stress-inducing conditions.

A full 24-h cycle was simulated, with four virtual IoT sensor nodes deployed in fixed positions across the area. Each node recorded environmental data every 60 s, specifically:

• Ambient temperature (°C).
• CO₂ concentration (ppm).

Temperature followed a sinusoidal diurnal pattern with a midday peak. To simulate heat stress conditions, the system introduced random spikes above 34 °C and increased CO₂ levels beyond 500 ppm between 11:00 AM and 3:00 PM. Additionally, 5% of the dataset includes randomly injected missing values to emulate data transmission faults. Timestamp jitter was added to simulate asynchronous sensor behavior.

Figure 2 presents a time series segment with highlighted critical events.

5.2. Scenario Description

This scenario models a shaded rest area within a rotational grazing system, typically used by cattle to avoid direct sunlight during peak heat hours. These zones are essential for animal comfort but can quickly become critical environments due to crowding and insufficient ventilation.

From approximately 11:00 AM to 3:00 PM, cattle tend to concentrate in these shaded areas, increasing ambient temperature through metabolic heat and elevating CO₂ levels due to respiratory accumulation. The limited airflow exacerbates the risk of thermal stress and discomfort.

By simulating this behavior, the experiment captures the kinds of microclimatic fluctuations that self-configurable IoT systems must detect and respond to. The scenario reflects real-world conditions where environmental risk can shift rapidly, underscoring the need for adaptive sensing in precision livestock management.

5.3. Model Evaluation Metrics

To assess the performance of the classifier embedded in each IoT sensor, we evaluated the model using a labeled test set derived from the synthetic thermal stress dataset. The Random Forest classifier was tested on 2400 samples, achieving high accuracy across both critical and non-critical zone classifications.

Table 2. Classification report for zone criticality detection.

Class	Precision	Recall	F1-Score	Support
Non-Critical (0)	0.98	1.00	0.99	2084
Critical (1)	0.98	0.85	0.91	316
Accuracy	0.98
Macro Avg	0.98	0.92	0.95	2400
Weighted Avg	0.98	0.98	0.98	2400

summarizes the model’s performance metrics, while Figure 3 presents the corresponding confusion matrix.

The results indicate a high precision for both classes, with particularly strong performance in correctly identifying non-critical conditions. The recall for critical conditions is slightly lower (0.85), reflecting a small proportion of false negatives (48 cases), which are visible in the confusion matrix below.

5.4. Critical Event Visualization

To better illustrate the system’s ability to capture environmental changes associated with thermal stress, Figure 4 shows a detailed time window from the simulated experiment.

The plot displays overlapping time series of ambient temperature (red line) and CO₂ concentration (green line) along with critical events marked by black cross symbols. These critical points correspond to sensor readings that exceed the defined thresholds for thermal stress: temperature above 34 °C and/or CO₂ levels above 500 ppm.

The figure clearly shows how environmental conditions rapidly shift in shaded resting areas, particularly around midnight when animal clustering and reduced airflow cause localized spikes in temperature and CO₂. The annotated critical events confirm the system’s sensitivity to these shifts, enabling timely adjustments in sampling rate and transmission behavior through the adaptive feedback loop.

6. Conclusions and Future Work

This work developed a self-configurable, IoT-based environmental monitoring framework designed for shaded resting areas in rotational grazing systems. The integration of embedded Random Forest classification and adaptive reconfiguration demonstrated that it is feasible to embed intelligence at the edge, allowing sensor nodes not only to capture environmental dynamics but also to autonomously adapt their operation to changing conditions.

The key outcome is that adaptive configuration at the sensor level substantially improves both responsiveness to critical states and energy efficiency—two competing requirements in remote livestock environments. This dual optimization moves beyond static monitoring by providing an infrastructure that is situationally aware and self-adjusting.

The findings also reaffirm the initial motivation: livestock farms require scalable, low-maintenance monitoring systems capable of early detection of suboptimal environments that compromise animal welfare. By linking detection accuracy (98% precision) with autonomous operational adaptation, our system closes the loop between monitoring and response.

In broader terms, this approach contributes to the digital transformation of livestock production, where resilience to environmental stress, reduced human dependency, and sustainable energy usage are essential. Future work will extend validation into real farm deployments, explore hybrid models combining edge and cloud intelligence, and assess long-term robustness under diverse grazing conditions.

Future work will involve deployment in real-world farm environments, extension to additional environmental and physiological variables, and integration with predictive decision support platforms to further support animal welfare and operational sustainability. Additional research directions include the implementation of meta-learning algorithms for rapid adaptation, autonomous analytical cycles for self-diagnosis and self-optimization, federated learning for privacy-preserving distributed modeling, and unsupervised anomaly detection for early warning in unknown scenarios. In addition, future work will integrate mitigation strategies such as ventilation, spraying, or shade control so that the system not only detects but also helps eliminate critical environmental conditions.