Handling Missing Value dengan Pendekatan Regresi pada Dataset Akuakultur Berukuran Kecil

Regression (SVR) 0.739, (KR) Abstract —Shrimp cultivation is strongly influenced by pond water quality conditions. Farmers must know the appropriate action in regulating water quality that is suitable for shrimp survival. The state of water quality can be understood by measuring pond parameters using various sensors. Installing sensors equipped with artificial intelligence modules to inform water quality conditions is the right action. However, the sensor cannot be separated from errors, so it results in not being able to get data or missing data. In this case, the approach of 5 parameters of pond water quality from 13 available parameters is carried out. This paper proposes a technique to obtain lost data caused by sensor error and looks for the best model. A simple approach can be taken, such as the Handling Missing Value (HMV) which is commonly used, namely the mean, with the K-Nearest Neighbors (KNN) classifier optimized using a grid search. However, the accuracy of this technique is still low, reaching 0.739 at 20-fold cross-validation. Calculations were carried out with other methods to further improve the prediction accuracy. It was found that Linear Regression (LR) can increase accuracy up to 0.757, which outperforms different approaches such as the statistical approach to mean 0.739, mode 0.716, median 0.734, and regression approach KNN 0.742, Lasso 0.751, Passive Aggressive Regressor (PAR) 0.737, Support Vector Regression (SVR) 0.739, Kernel Ridge (KR) 0.731, and Stochastic Gradient Descent (SGD) 0.734.