Ted Talk Views Prediction

1. Objective: To build predictive model, which help in predicting views of video uploaded on TEDX website.

1.1 Data overview:

Number of TED talks: 4,005
Number of speakers: 3,274
Number of events: 459
Timespan: Jun 2006 - Apr 2020
Attributes: 19

1.2 Type of Machine learning problem

Task is to predict views of TED talk videos => Regression problem

1.3 Performance Metric

1. MAE,MSE,RMSE
2. R2_Score

2. Exploratory Data Analysis

2.1 Distribution of TEDX videos by weekdays with sum of target variable (views)

2.1 Distribution of TEDX videos by weekdays with sum of target variable (views)

3. Feature Engineering

• Multiple Speakers in Ted Talk we have converted this feature with speakers_avg_views, Mutiple events converted to event_avg_views etc etc
• Year feature has been converted to Age of video with (2023-(Year when video was published)) to understand how much older the video is.
• Used One hot encoding on Orignal langauge in which Ted Talk was released to compare Videos langauge wise.

4. Feature Selection

4.1 f_regression to get feature importance, Dropping features with higher P-value with threshold > 0.4

4.2 Correlation

Observation: comments_end_distribution is 77% co-rrelated with comments_NaN DROP Comments_NAN

5. Comparison of R2_Scores on diffrent Machine learning Models used.

1. Linear Regression

2. Ridge Regressor

3. Lasso Regressor

4. Random Forest Regressor

5. XGB Regressor

After comparing all the models w.r.t.o their R2 scores even simple models like linear & Ridge can perform better than ensemble models like Random Forest and XGB with this type of use cases.