Microbusiness Density Forecasting Based on XGBoost

Abstract

Modern American business culture heavily depends on microbusiness. Defined as a specific type of small business with an online presence and no more than ten employees, microbusiness faces certain challenges that distinguish it apart from its larger counterparts, including a lack of financial resources, sh279279ort of assistance and concerns from authorities. Recognizing the critical need to understand and support this vital segment of economy, the study will employ advanced machine learning methods to predict the next month’s microbusiness density based on historical time series data set. Specifically, the utilization of XGBoost algorithm is coupled with appropriate feature engineering to create a robust predictive model. Delving into the algorithm structure, anomalies are identified and corrected through data smoothing, utilizing rolling window sum features to capture recent trends, and the process of creating lag features. Manually filtering out anomalies by setting boundaries, this study fits the model using the constructed features to make adequate predictions. On microbusiness prediction, multiple forecasting models are compared and their performance is assessed using a test set that is well-known. The result presents that XGBoost algorithm outperforms traditional time series models when predictions are solely based on previous observations. The approach of the research emphasizes the significance of leveraging cutting-edge machine learning techniques to gain insight into the microbusiness landscape, thereby enabling more targeted support and opportunities for this crucial sector of the U.S. economy.