Kalman Filter Method for Handling Missing Values in Soil Moisture Sensor Data

Abstract

Several imputation techniques have been developed specifically to deal with missing values. This research used data from soil moisture sensors for 34 days where there are missing values. Size of soil moisture sensor data to be quite large. So that with the missing value, it is difficult to determine how well the imputation technique is applied. Therefore, imputation technique is performed on the generated data based on the distribution of soil moisture sensor data so that an evaluation of the utilization of the imputation technique can be carried out on large data containing missing values. The method used in this study is Kalman Filter. Evaluation using the Mean Average Percentage Error (MAPE) after intentionally removed using the Missing Completely at Random (MCAR) technique with a missing rate of 5%, 10%, and 15%. The results showed that soil moisture data has a Box-Cox power exponential distribution. It was found that for generating data, Kalman Filter method has not much different MAPE value for 100 and 1000 data with missing rate was 5%, 10%, and 15%. The results of estimating missing values with the Kalman Filter on the soil moisture sensor data are in line with the soil moisture sensor data