The purpose of this paper is to explore and develop an analysis algorithm of sports performance data suitable for military skill teaching, so as to improve the training effect and military skill level of soldiers. Traditional teaching methods rely on the subjective judgment of instructors and lack scientific and objective evaluation criteria. Therefore, this article proposes a deep learning-based data analysis algorithm that combines convolutional neural networks (CNN) and long short-term memory networks (LSTM) to process and analyze soldiers’ motion performance data during military skill training. Algorithm design includes four main steps: data preprocessing, feature extraction, model training, and evaluation. The data preprocessing stage cleans and formats the data to ensure data quality. In the feature extraction stage, principal component analysis (PCA) is used for feature dimensionality reduction, and key features related to soldier movement performance are extracted. In the model training stage, CNN-LSTM model is trained by using these features to realize the classification and prediction of soldiers’ sports performance. The experimental results show that the CNN-LSTM model proposed in this paper performs well in predicting soldiers’ sports performance, with an accuracy rate of 0.95, which is significantly higher than traditional machine learning models such as logistic regression, support vector machine (SVM) and random forest. Through confusion matrix analysis, the accuracy of the model in identifying the extreme categories (excellent and poor) of soldiers’ sports performance reached 88.9% and 83.3% respectively. In addition, the algorithm can feed back the soldiers’ sports performance in real time, and help instructors and soldiers find problems in time and make targeted improvements. The research in this paper not only provides a scientific and objective evaluation method for military skill teaching, but also provides a new idea for the application of sports performance data analysis in the military field.