Development of a Funded Defense Companies Prediction Model based on Machine Learning

This research focused on the needs of the defense industry to understand what companies should preemptively prepare from the perspective of their current management status in order to increase the probability of receiving benefits from being selected for defense industry development support projects. An experiment was conducted to build a prediction model for beneficiaries of the parts localization development support project and the weapon system modification development support project using corporate information. To compensate for the imbalance problem of data classes of variables, random sampling methods such as oversampling, undersampling, and hybrid sampling methods were applied. The backward elimination technique was applied as a variable selection technique, and the ensemble technique was additionally applied to improve the performance of the prediction model. What differentiates this study is that it analyzes corporate characteristics that determine beneficiary and non-beneficiary companies and builds a prediction model for beneficiary companies.