This article presents a clustering method that concentrates on statistical noise reduction techniques utilising the Wavelet transform (WT) and evaluation to circumvent the challenges of current patterns in handling the non-stationary and non-linear characteristics of high-frequency critical time-domain records, specifically their poor generalizability possibility. Utilising the Neuronal Society of Fast Long-Term Short Memory (LSTM) and singular spectrum analysis (SSA), a model for information prediction is constructed. To avoid overemphasising learning, an early preventive strategy is included into the educational process after studying the stock data of the old-time group and building the time collection with special days based on community participation. Very fitting. As a last step, we utilise the state parameter transfer technique and the variable term batch to predict the remaining inventory charges on the test set. Our results show that the LSTM prediction model an