Firm Order Estimation – Combining Artificial Intelligence and Firm Data to enable Enhanced Forecasting and Key Performance Indicator Estimation

Abstract

This study investigated the performance of K-Nearest Neighbors (KNN) and Long Short-Term Memory (LSTM), a Recurrent Neural Network variant, in estimating missing data in both synthetic and real firm databases. Our objective was to identify the most effective model for imputation, considering estimation accuracy and robustness to data-specific characteristics, such as outliers and noise. Although KNN showed superior performance based on Mean Absolute Error and Mean Squared Error metrics, high Mean Absolute Percentage Error values observed for both models suggest potential issues, such as overfitting and the influence of extreme values. Neither model demonstrated significant promise for missing data estimation in this context, emphasizing the need for careful data preprocessing, model selection, and parameter tuning. Consequently, future research should consider alternative preprocessing techniques and machine learning models, underlining the importance of a nuanced understanding of the data and careful model and parameter selection for robust and accurate prediction outcomes.