The implementation is about predicting market-share about some some episodes in Montreal Local TV Channels between 2016-2019
These data belong to the Montreal Local TV channels, which contains several features including: Name of episode, data, time, etc.
Dataset has around 12 feature columns and about 600k samples.
Pandas: Used for data manipulation
Numpy: Powerful library for numerical computation
Matplotlib: Useful in plotting and data visualisation
Scikit-learn: Powerful library with implementation most of the machine learning library and performance metric
joblib: for saving the trained model.
Montreal_TV.py: A python script for model training, saving the model and generate prediction.csv
Montreal_TV.ipynb: A jupyter notebook with additional data visualization and analysis.
Random Forest Regressor: After testing several models including: SGDRegressor, Lasso, Decision Tree, Random Forest it is found that RF is best suitable for this task.
Metrics were MAE and R-Squared.
| Metric | Train | Test |
|---|---|---|
| R-Squared | 0.91 | 0.88 |
| MAE | 0.96 | 1.05 |
1-Scikit-learn pipelining: by using Scikit-learn pipelining, data cleanning can be done better and code readability enhances.
2-Apache Spark Distributed Tools: by using apache spark speed of the training phase may be increased by help of distributed processing.