Netflix Popularity Prediction

Predicting pre-release success using metadata features, cast/director ratings, and natural language plot profiles.

Open repository

Project claim

Predict pre-release content success using synopses and metadata.

Role

Machine Learning Engineer

Focus signals

80.6% accuracy0.721 ROC-AUCTF-IDF plot keywords

Proof signal 1

80% Random Forest accuracy

Proof signal 2

TF-IDF synopsis predictive power

Proof signal 3

Power BI dashboard integration

Challenge

Identifying pre-release content success factors before expensive production investments.

Solution

Engineered features for director track records, genre popularity, duration, and textual plot synopsis themes (TF-IDF), training Random Forest classifiers.

Build notes

Handled class imbalances using advanced pre-processing and stratified data splits

Used TF-IDF and NLTK stopword filters to clean synopses descriptions

Built correlations visualizations and correlation heatmaps to assess feature importance

Tools + stack

Python, scikit-learn, Random Forest, and Power BI.

Pythonscikit-learnNLTKPower BIRandom ForestTF-IDF

Overall accuracy

80.60%

ROC-AUC

0.721 score

Held-out test set

20% partition

Architecture flow

Step 1

netflix dataset

Step 2

synopsis text

Step 3

nlp feature prep

Step 4

random forest

Step 5

power bi logs

Predictive NLP

Extracted vocabulary signals from text metadata, analyzing plot descriptions to identify keywords that strongly correlate with audience success.

Feature Importance

Found duration, release year, and specific plot keywords had higher relative feature importance than traditional markers like cast popularity or budget.

Decision signals

Intrinsic content attributes (plot keywords, duration) hold more predictive power than financial budgets

A baseline model (Logistic Regression) can offer higher recall, but Random Forest yields superior precision

Feature dependencies require correlation mapping before training model ensembles

Outcomes

Trained Random Forest classifiers yielding 80% accuracy, outperforming Logistic Regression baselines

Extracted predictive keywords ('finds', 'help', 'young') from plot synopses, showing semantic influence on show popularity

Developed dynamic Power BI dashboards representing KMeans clustering partitions and correlation heatmaps