Machine Learning and Wavelet Transform: A Hybrid Approach to Predicting Ammonia Levels in Poultry Farms.

Area of Science:

• Agricultural Science
• Environmental Science
• Data Science

Background:

• Ammonia (NH3) is a significant poultry farm pollutant impacting bird health and industry economics.
• Accurate NH3 monitoring is essential for environmental protection and animal welfare.

Purpose of the Study:

• To evaluate machine learning algorithms for predicting ammonia concentrations in poultry farms.
• To assess the effectiveness of Wavelet Transform as a preprocessing technique for NH3 prediction models.

Main Methods:

• Three machine learning algorithms (Extreme Learning Machine, K-Nearest Neighbor, Random Forest) were tested.
• Wavelet Transform with ten decomposition levels was used as a data preprocessing step.
• Predictive accuracy was evaluated using Mean Absolute Error (MAE) and correlation coefficient (R).

Main Results:

• The Random Forest (RF) algorithm demonstrated robust performance, both independently and combined with Wavelet Transform (WT).
• The RF-WT model achieved the best prediction accuracy using air temperature, relative humidity, and air velocity.
• The optimal RF-WT model yielded a MAE of 0.548 ppm and an R of 0.976 on the testing dataset.

Conclusions:

• Preprocessing poultry farm data with Wavelet Transform significantly enhances the predictive power of machine learning models for ammonia.
• The Random Forest-Wavelet Transform combination offers a highly accurate method for estimating ammonia levels, crucial for farm management and environmental control.