Data Completion for Model Success Execution using Incremental MLP and LSTM Models

Almost all the models is the network today is running on different datasets, including the market analysis, demand forecasting, and healthcare policy development. However, missing or incomplete data, caused by factors such as data loss or market fluctuations, poses a significant challenge for downstream tasks. Traditional methods, like mean imputation, often fail to capture complex price variations. This paper proposes a drug price imputation method using Incremental MLP (IMLP) and LSTM models. We first clean the missing drug names and prices and apply normalization to handle price variations. Then, we use One-Hot Encoding to convert drug names into numerical vectors, forming the feature matrix, which is eventually input into the MLP and LSTM models. In addition, we introduce the IMLP model, which incorporates incremental learning to adapt to new drug types and price changes. Experimental results on four healthcare insurance drug price datasets show that IMLP outperforms LSTM, achieving an average R2 0.48 higher and 36.1% lower Mean Squared Error (MSE).