Prediction of Company Financial Performance From Financial Statements and Stock Price Using LSTM

The Financial performance of a company is important for investors and Board of Directors also employees who run the business. Investors and those running the company want the company to succeed, get investment growth for investors and increased compensation through effective business operations for the Board of Directors and employees. A company's financial performance is reflected in its financial statements.especially for companies listed on the stock exchange where the share price also reflects the company's performance. Evaluation of financial statements and stock prices can be done with Value Investing, a method that determines the fair value of a stock. the method using financial parameters and ratio with the stock prices. This research collecting financial statements and stock prices form Indonesia stock exchange. Data put from 2 companies from bank sector there are BBRI (Bank Rakyat Indonesia), and BMRI (Bank Mandiri) and 2 companies from energy sector there are PGAS (Perusahaan Gas Negara/ Pertamina Gas), and INDY (Indika Energy). From the data of 4 companies, the financial statements and stock price data are taken and normalized using min - max scaling, standart scalling and Robust Scaling, then trained using LSTM, Random Forest Regressor and SVM (Support Vector Machine). The performance results of the models are compared for each data normalization and machine learning model. The Random Forest Regressor has an overall more stable and consistent performance than LSTM and SVM, with lower RMSE and MAPE values across different scaling scenarios. LSTM tends to be more variable in performance, depending on the scaling scenario and stock type, but still shows potential in certain stocks with Robust Scaling. SVM shows more fluctuating performance, especially with Robust Scaling, which leads to significant errors on some stocks. The use of MinMax Scaler often resulted in lower errors than other scaling techniques, especially for Random Forest and SVM models.