Predictive modelling

Hydrogen sulfide

Engineering

Statistics

Support vector machine

Computer science

Odor

Laying

Organic chemistry

Mathematics

Environmental science

Artificial intelligence

Sulfur

Structural engineering

Chemistry

Machine learning

Computers and Electronics in Agriculture

In laying hen farm, odor measurement and reduction are necessary for clean environment management and human/animal health care. However, limited quantitative data and expensive detection technologies preclude an accurate assessment of odor reduction practices. This study compared the odor prediction ability of three different machine learning models of extreme gradient boosting (XGBoost), support vector regression (SVR), and back-propagation neural networks (BPNN) based on small sample size odor datasets collected from laying hen houses to achieve timely odor monitoring and check the significant components affecting the odor concentration. The input variables were ammonia (NH3) concentration, hydrogen sulfide (H2S) concentration, temperature, relative humidity (RH), and ventilation rate, and the output value was odor concentration. Results showed the XGBoost model had the highest prediction ability and the most accuracy with the R2 of 0.88, followed by BPNN (R2 = 0.75) and SVR (R2 = 0.66). XGBoost model could be a useful tool to timely predict odor concentration with moderate accuracy. In addition, in the trained XGBoost model, NH3 concentration was the most important factor, followed by the H2S concentration, temperature, RH, and ventilation rate, which indicated the gas components and environment variables related to gas production were the key drivers in training odor prediction model. This study also forecasted the influences of each gas and environmental factors on odor concentrations and verified the good knowledge mining ability of XGBoost model. The practical potentials are considerable of using XGBoost model to replace human assessors and to apply to laying hen house environmental control as the odor sensor.

Prediction of laying hen house odor concentrations using machine learning models based on small sample data