Using Optimizing Parameters Support Vector Regression Model to Predict Potassium Ratio in Carb Fish

Abstract

In this paper, we studied the combination of the levels of carbohydrates (20%, 30%, 40%, and 50%), protein (8%, 12%, 16%, and 20%), and fats (5%, 10%, 15% and 20%), where all possible combinations are 64. We gave each combination of the aforementioned elements to an aquarium fish with a volume of 1.92 m3, each aquarium contained 5 fishes, the aim of our study is to detect which combination of the three elements recorded a high potassium ratio of the fishes, here we depend on the average of the fishes weight and the results clarified that the combination (30%, 8% and 15%) of 1kg for carbohydrate, protein, and fats respectively are given average potassium ratio of 503mg/kg, for this purpose optimized parameters SVR has been used. According to the results radial kernel function with optimized parameter (gamma = 1.341 and cost = 0.844) gave the highest performance compared to the other kernel functions, the R2 = 91% this implies the factors capable of explaining 91% of fishes weight with MSE and RMSE of (0.000438 and 0.02092) respectively. And p-values of the three aforementioned variables are less than the significant level of 0.01, implying that the three factors have a statistically significant impact on the fish’s weight. Where carbohydrate has an impact of 0.12 on the fish’s potassium ratio, in another word if carbohydrate increase by one unit, then the fish’s potassium ratio increases by 0. 12 mg, also both protein and fats have a significant positive effect on the response variable, and the amount of impacts are (1.015 and 0.117) respectively [13].