Performance Comparison of Data Preprocessing Methods for Trade-Space Exploration with AI Model: Case Study of Satellite Anomalies Detection

Satellites are critical components of modern infrastructure, supporting countless applications in communication, navigation, and observation. However, ensuring their functionality and safety within complex space environments can be challenging. The satellite experiences the highest loss in the space industry caused by anomalies. Hence, it needs early detection so that the loss can be avoided immediately. With the advancement of technology, satellite anomalies diagnosis and detection can be done with trade-space exploration (TSE) and Artificial Intelligence (AI) models based on satellite data. The problem is that in satellite data preprocessing step, the data can be too large and sometimes there are some missing values encountered which leads to outliers. To mitigate these problems, efficient data preprocessing is needed so that the accuracy can be leveraged and requires only minimal computation resources. This paper presents the examination of the data preprocessing performance from the combination of both data cleansing and data normalization methods. Elimination, Imputation, Feature of Missing and Imperative Imputation methods are involved in data cleansing. While for the data normalization presented, Min Max, Z-Score using Standard Scalar, Robust Scaling, Vector Normalization and Power Transformation methods are used. As for the AI model classification, it is using Support Vector Machines (SVMs). The test was conducted using data from Satellite Database and Space Market Analysis (Seradata) consisting of approximately 4,455 data. The result shows that the accuracy of the Elimination and the Power Transformation normalization is the highest in training accuracy with 60%. While the Elimination and the Min Max or the Z-Score methods are the top in the testing accuracy with 60%.