Ozone Day Prediction Using a Combination Method of Matrix Completion and Interactive Lasso

Abstract

The missing data classification problem is one of the common problems in machine learning. Conventional method eliminates the samples with missing values. In this paper, matrix completion, as a new method is proposed for filling the missing data. And this method and two traditional methods, eliminating the samples with missing values and filling the missing data based on the sample similarity, are compared through experiments on the ozone classification data. In addition, the ozone day prediction depends on complex interaction information among data features, so the interactive lasso model is proposed for interaction feature selection and classification. The interactive lasso method is compared with the lasso and random forest (RF) methods. The final experimental results demonstrate our combination method. The classification accuracy of ozone day is approaching 100%.