A New Method for Handling Large Missing Values from Rain Gauge Sensors

Many Internet of Things (IoT)-based systems transmit data quickly in a short time. Unfortunately, not all IoT devices transmit complete sensor data, resulting in data loss, which is a common problem faced by IoT-based systems. Additionally, not all data sent by sensors exhibits a strong correlation between variables. Rain data sourced from the Meteorology, Climatology, and Geophysics Agency (BMKG), which is derived from rain gauge sensors, is an example of a data variable with a very low correlation with other variables, such as temperature, humidity, and wind, and can therefore be considered a univariate variable. Solving the problem of missing values with univariate data characteristics is undoubtedly different from the attributes of multivariate data. Therefore, an appropriate imputation method is required that can address this issue. This study proposes a method that results from the fusion of multiple imputation and interpolation to impute live stream data of univariate variables. The proposed method is then compared with several single imputation methods commonly used for data imputation, such as mean and median imputation. The proposed method yields significantly better results compared to the mean and median imputation methods, particularly when the percentage of missing data exceeds 30%. Based on these results, the problem of missing values with a large percentage in real-time univariate data can be solved by this method. Thus, if a dataset contains no missing data, then prediction and other processes can yield better accuracy.