Analyzing Social Media Posts with Python: From Data Cleaning to Machine Learning Prediction

This article introduces a complete social media data analysis project, showing how to use Python to go from raw data through the entire workflow of data cleaning, visualization, cluster analysis to machine learning prediction (regression and classification), helping to extract valuable insights from social media posts. The project is from the Big-Data-Analyst-for-facebook project on GitHub, authored by Abdelrahman Ali, published on May 31, 2026.

The core goal of the project is to understand the performance patterns of social media posts and analyze interaction data (likes, shares, comments, impressions, reach, etc.). The dataset used, data_15.csv, contains multi-dimensional information such as post type, category, publication time, reach, impressions, engagement, number of likes, comments, shares, and total interactions, providing a foundation for in-depth analysis.

Data preprocessing includes: 1. Missing value handling (numerical columns filled with median, non-numerical columns filled with mode); 2. Removing duplicate rows to ensure data purity; 3. Categorical variable encoding (LabelEncoder for Type and Category, one-hot encoding for Post Weekday); 4. Min-Max normalization of numerical columns to the 0-1 range to improve algorithm performance.

Data patterns are revealed through various charts: histograms show the distribution of numerical features, box plots detect outliers, scatter plots present the relationship between likes and shares, correlation heatmaps show the degree of variable correlation, and pair plots discover multi-variable interaction patterns. These visualizations help understand post propagation mechanisms and optimize content strategies.

Cluster analysis uses K-Means and hierarchical clustering, with 3 clusters determined by the elbow method, using features such as reach and number of engaged users; regression models predict total lifetime reach (evaluated by MSE); classification models use Random Forest and SVC to predict post type (evaluated by accuracy and classification report).

The project demonstrates the complete lifecycle of data science, which is of reference value to learners and practitioners. The analysis results can optimize content strategies to improve propagation effects. The tech stack uses mainstream Python libraries (pandas, numpy, matplotlib, seaborn, scikit-learn), and the code is well-organized for easy reproduction.