Performance Evaluation of Predictive Models for Missing Data Imputation in Weather Data

Real datasets can have missing values for a different reasons such as in data that were not kept on file and data corruption. Climate forecasting has a highly relevant effect in agricultural fields and industries sectors. The process of predicting climate conditions is required for different areas of life sectors. Handling missing data is significant because a lot of machine learning algorithms performance are affected by missing values in addition, they do not support data with missing values. Various techniques have been used to process missing data problem and the most applied is removing any row that contains at least one missing value. Also, another approaches to solve missing data problems are to impute the missing data to yield a more complete dataset. In order to improve the accuracy of prediction with the climate data, missing value from dataset should be removed or imputed/predicted in the pre-processing phase before using the data for prediction or clustering in the analysis step. In this paper, we propose a new technique to handle missing values in weather data using machine learning algorithms by execute experiments with NCDC dataset to evaluate the prediction error of five methods namely the kernel ridge, linear regression, random forest, SVM imputation and KNN imputation procedure. The missing values were imputed using each method and compared to the observed value. Results of the proposed method were compared with existing techniques.