Now-a-days demand forecasting is used in many countries for military applications such as spare parts of aircraft and for improving budget efficiency. In supply chain management demand forecasting is a major issue. Currently, time series technique is used to demand forecast but this technique resulted in lack of accuracy and improvement in accuracy is needed. So this paper focused on comparing the features which leads to improvement in the accuracy and propose a system for demand forecasting of spare parts of anti-aircraft missiles, which are based on machine learning and neural networks such that equipment's are properly utilized and alongwith that budget is also maintained. We have compared the existing features with the new features added and applied algorithms and looked upon at the accuracy. Experimental results proves that the new features added gave higher accuracy. Here, we also present an end-to-end boosting system called XGBoost and Multi-layer Perceptron. These new techniques were compared with traditional Machine Learning techniques. This experiment is conducted on the Vietnam War dataset. © 2019 IEEE.