Predicting Reach and Engagement for a Programming/Data-Science Educational Instagram Page Using Regression Models

Social media performance prediction helps content creators and educational pages understand which posts are likely to receive higher reach and engagement. This study focuses on predicting the reach and engagement of a programming/data-science educational Instagram page using historical post metrics, captions, and hashtags. The raw dataset contains 119 Instagram post records with 13 columns. After removing 17 exact duplicate rows, 102 unique posts are used for analysis and modeling. The task is formulated as supervised regression because the output variables are numeric counts. Four target variables are considered: Impressions, Likes, Comments, and Shares. The study compares two feature views: content-only features, which are available before publishing a post, and performance-signal features, which include traffic-source and account-response variables that are available only after posting or must be estimated. Models evaluated include a mean baseline, Ridge regression, Random Forest, and Gradient Boosting. Results show that content-only features provide weak predictive performance on this small and narrow dataset, while performance-signal features substantially improve prediction, especially for Impressions. The best performance-signal model achieved a holdout R2 of 0.998 for Impressions, 0.589 for Likes, 0.181 for Comments, and 0.458 for Shares. The main contribution of this work is the clear separation of pre-publication prediction from performance-signal prediction, preventing misleading claims about model accuracy. The study is best framed as an account-specific exploratory prediction system rather than a general Instagram popularity prediction model.