Click Count Forecasting with Bi-Directional LSTM and Temporal Feature Fusion

In recent years, online advertising has become a fundamental element of the digital economy, with businesses increasingly relying on digital ads to reach potential customers and boost sales, where key metrics like impressions, clicks, click-through rates (CTR), and conversions are used to measure effectiveness. Among these, click counts-representing the frequency of user interactions with an ad-are crucial for evaluating campaign success, and accurately predicting these counts is essential for optimizing advertising expenditures and enhancing user experiences by delivering relevant content at the right time. This research investigates the application of Bi-Directional Long Short-Term Memory (Bi-LSTM) networks with temporal feature fusion to predict click counts in online advertising, presenting a notable advantage over conventional models by capturing non-linear relationships in the data and detecting complex patterns through the integration of past and future temporal features. The proposed model improves prediction accuracy by better understanding the context surrounding ad views, with the results demonstrating that incorporating interaction terms between features leads to superior performance, achieving a Mean Squared Error (MSE) of 2.73, a Root Mean Squared Error (RMSE) of 1.65, a Mean Absolute Error (MAE) of 1.37, and the highest R2 score of 0.88, which highlights the model's ability to account for complex relationships within temporal data.