Social Media Views Predictor (LightGBM Quantile Regressor)

Model Description

This repository contains two machine learning pipelines designed to predict the number of views for social media posts (Facebook & Instagram) based on engagement metrics and content type.

  • Algorithm: LightGBM Regressor (Objective: Quantile, alpha=0.5).
  • Why Quantile Regression?: Social media views are highly skewed with extreme outliers (viral posts). Predicting the median (quantile 0.5) instead of the mean makes the model significantly more robust to these extreme outliers, minimizing standard error inflation.
  • Preprocessing Pipeline:
    • RobustScaler: Applied to numerical engagement features (Reactions, Comments, Shares, Total Engagement) to mitigate the impact of engagement outliers.
    • OneHotEncoder: Applied to categorical features (Source platform, Content Type).

Files in Repository

  • best_model.joblib: The model fine-tuned using standard 5-fold cross-validation.
  • best_model_ts.joblib: The model fine-tuned using TimeSeriesSplit (chronological cross-validation) for strict temporal evaluation.

Intended Use

Predicting expected median views to help creators and marketers understand expected baseline performance before a post goes highly viral.

Format

The models are saved as joblib files, which is the recommended and most efficient format for scikit-learn pipelines containing NumPy arrays.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support