<![CDATA[The development of digital technology has increased online purchases, changed consumption patterns, and created a need for accurate sales predictions for strategic decision-making. However, prediction accuracy is often hindered by multicollinearity issues among numeric features and the risk of overfitting in conventional linear regression models. This research aims to address these issues by applying Ridge Regression based on L2 regularization to predict total sales using historical e-commerce data. The methods used include data collection (541,909 records), preprocessing (removal of outliers and missing values), and modeling with scikit-learn, evaluated using MAE (73.25), RMSE (129.80), and the coefficient of determination (R²). The results indicate that the model is capable of predicting total sales with adequate accuracy even though there are deviations in extreme value transactions. Practically, the implementation of Ridge Regression has proven effective in reducing overfitting and enhancing prediction stability, making it a reference in inventory planning and marketing strategies. This finding reinforces the need for adaptive machine learning approaches in sales data analysis in the digital era.]]>
Copyrights © 2026
