The Multicollinearity Effect on Machine Learning Accuracy for Leaf Chlorophyll Content Prediction of Indoor Plants

Various factors that influence chlorophyll levels in indoor plants were analysed. The input dataset consisted of 52 samples, which represent 10 distinct plant types. To non-invasively measure the chlorophyll content in plant leaves, Soil-Plant Analysis Development (SPAD) sensor was used, measuring the absorbance of specific light wavelengths, allowing for the assessment of chlorophyll concentration. The dataset was supplemented by covariates from soil electrical conductivity (EC) sensing at depths of 5 cm, 10 cm, and 15 cm, along with the multispectral Plant-O-Meter sensor. Four covariates in the model, including plant type, EC (5 cm), EC (15 cm), and normalized difference red-edge index (NDRE), showed minimal correlation with other variables, highlighting their independence. To predict leaf chlorophyll content, Random Forest and Extreme Gradient Boosting machine learning models were utilized, with Random Forest achieving higher average coefficient of determination of 0.458. The study underscored the potential of a complementary dataset for evaluating the complex relationship among root-soil dynamics and leaf for optimizing indoor plant health.