Integrated allometric scaling, physicochemical profiling, and random forest modelling for predicting antioxidant compounds in sweeteners from overripe Malaysian bananas

Abstract Bananas ( Musa spp. ) are rich in bioactive compounds that are essential for health, including support for digestion, immunity, and sustained energy. This research compared the nutritional profiles, physicochemical properties, determinants of antioxidant compounds, and antioxidant activities of sweeteners from eight overripe banana species ( Tanduk, Batu, Nangka, Emas, Cavendish, Awak, Lemak Manis, and Kapas ) in Malaysia, using validated methods. All the chosen sweeteners produced from the different banana species exhibited significant differences in physicochemical and antioxidant properties. Based on the results obtained, Tanduk exhibited the highest TDF at 8.89 g/100 g, Lemak Manis exhibited the highest ash content (4.05 g/100 g) and crude fiber (0.80 g/100 g), Kapas exhibited the highest carbohydrate content (71.99 g/100 g), and Emas revealed the highest protein content (0.62 g/100 g). Conversely, Nangka exhibited the lowest antioxidant capacity but the highest sugar content, especially glucose. Cavendish exhibited the highest antioxidant capacity, as evidenced by the DPPH and FRAP assays. The Random Forest model identified banana species, caloric content, carbohydrates, total dietary fiber, peel weight, protein, colour L*, pulp peel ratio, and crude fibre as the primary determinants of variation in antioxidant compounds among the sweeteners of Malaysian banana species. The power-law model demonstrated a strong allometric relationship between pulp weight and fruit weight. Future work could integrate feature selection techniques and explainable AI methods to further enhance biological interpretability by quantifying the contribution of each nutritional and physicochemical attribute to antioxidant activity.