An Investigation into the Prediction of Annual Income Levels Through the Utilization of Demographic Features Employing the Modified UCI Adult Dataset

Predicting an individual's annual income is crucial and has many uses for the planning of government, policymakers, and development partners. This paper uses the UCI Adult Dataset to examine demographic-based annual income predictions. The 16-column dataset classifies “Income” into two categories: $< =50\mathrm{K}$ and $> 50\mathrm{K}$, where 14 characteristics provide personal information. Appropriate Methods to Handle Categorical Data guidelines govern the study's preprocessing, categorical data management, and missing value strategies. Eleven machine learning models are then studied, including classification, clustering, neural networks, Logistic Regression, Naive Bayes (standard and Gaussian), K-Nearest Neighbors, SVMs, Decision Trees, Random Forest, XGBoost, and Artificial Neural Networks. This study emphasizes optimization, evaluation, and model comparison and how optimization strategies and parameters affect model performance. Comparative analyses compare model performance within a category. The results imply that demographic characteristics help some models predict income levels better. Logistic Regression provides essential knowledge; whereas XGBoost and Neural Networks improve accuracy, Clustering adds knowledge. This study's findings inform revenue-anticipation-related decision-making with an accuracy of 87% from XGBoost and XGBoost-ANN Ensembling. This study emphasizes the need for data pretreatment, algorithm selection, optimization, and evaluation for accurate predictions and valuable insights.