ARCHIVES

This research explores a data-driven alternative to traditional machine learning experimentation by utilizing publicly available datasets instead of conducting resource-intensive and time-consuming physical studies. The objective is to investigate the predictive performance and interpretability of multipleregression algorithms, namely Linear Regression, Ridge Regression, Lasso Regression, and Decision Tree Regression, on real-world data. These models are evaluated based on their ability to uncover meaningful patterns, relationships, and potential causal inferences within the dataset. Emphasis is placed on essential preprocessing steps, including data cleaning, transformation, and encoding, to ensure the quality and consistency required for reliable model training and evaluation. By comparing these algorithms across common performance metrics such as mean squared error, mean absolute error, and R² score, the study provides insights into their robustness and generalizability. This approach facilitates repeatable and scalable experimentation, supporting rapid hypothesis testing and model optimization without the logistical constraints of physical data collection. Overall, the methodology contributes to accelerating machine learning research by promoting efficient, cost-effective practices while maintaining scientific rigor.