Predicting Instagram Post User Engagement With Machine Learning Models

Abstract

This study investigates the practical application of statistical machine learning techniques to predict average Instagram post user engagement based on an account’s follower count. It explores three distinct models—linear Regression, Random Forest Regression, and Neural Networks—for their effectiveness in modeling engagement patterns. Using recent Instagram data, each model is trained on the average user engagement for a profile. The predicted data is then compared to the actual data to determine the most accurate and viable model. The Neural Network excelled in capturing variance, having the highest R-squared value of the three models tested, but struggled with overfitting. Random Forest handled non-linear patterns well, having the lowest mean squared error out of the three modes, but it tended to overestimate. LASSO Regression was a balance between both the Neural Network and Random Forest Model, maintaining variance capture while reducing overestimation. Future research could refine models or explore hybrid approaches for better scalability. Machine learning shows promise in predicting post popularity, but further improvements are needed to aid social media creators and developers.