Research on Digital Cost Dynamic Monitoring Algorithm Based on Deep Learning

This study proposes an A-LSTM algorithm that integrates attention mechanism and long short-term memory network (LSTM) for dynamic monitoring of digital costs. In the data preprocessing stage, the principal component analysis (PCA) and mutual information analysis are combined to deeply explore the potential correlation characteristics of cost data. The experiment uses tens of thousands of cost operation data from a large manufacturing enterprise, which are divided into training set, validation set and test set according to 70%, 15% and 15%. After comparative experiments, the prediction accuracy of the A-LSTM algorithm on the test set reached 95%, far exceeding the 80% of the traditional ARIMA algorithm and the 88% of the ordinary LSTM algorithm; the recall rate reached 92%, and the F1 value was 93.5%, with significant comprehensive performance advantages. Through visual analysis such as the cost prediction value and the actual value comparison line chart, the abnormal cost identification confusion matrix, etc., the A-LSTM algorithm can clearly identify cost anomalies and accurately predict cost trends, providing an efficient and accurate digital monitoring solution for enterprise cost management.