Machine learning powered prediction of photodegradation of 2,4-dichlorophenoxyacetic acid using gold-doped bismuth ferrite
Published in Cleaner Water, 2025
Recommended citation: Ishtiaq, R., Jaffari, Z. H., Abbas, A., & Lam, S.-M. (2025). Machine learning powered prediction of photodegradation of 2,4-dichlorophenoxyacetic acid using gold-doped bismuth ferrite. Cleaner Water, 4, 100163. https://doi.org/10.1016/j.clwat.2025.100163
2,4-D is among the world’s most heavily used herbicides and it persists in wastewater, where one route to destroying it is to shine light on it in the presence of a catalyst. Testing every combination of catalyst recipe and operating condition at the bench is impossibly slow, so 1,050 measurements on bismuth ferrite catalysts — including versions seeded with gold — were pooled and used to train models predicting how much herbicide each combination destroys. Weighing fourteen variables against one another, the models found that how the process is run mattered more than the fine chemical make-up of the catalyst itself. The study also reports the settings that gave the highest breakdown: a catalyst dose of 1.5 g/L, a light intensity of 105 W, neutral acidity, and a starting herbicide concentration of 5 mg/L.