Supervised Machine Learning algorithms are the most widely used. With this model, the data scientist acts as a guide and inserts labels that guide the algorithm. Just as a child learns to identify fruits by memorizing them in a picture book, in Supervised Learning the algorithm is trained on a dataset that is already labeled and has a predefined output. Examples of Supervised Machine Learning include algorithms such as linear and logistic regression, multiclass classification, and support vector machines. We also have Unsupervised and Reinforcement Algorithms with specific applications.

We use the LGBMRegressor (Multiple Regression Algorithm).

We use YouTube data for training and testing the algorithms. Since this platform holds a significant market share, we were able to cover a wide spectrum of the various commercial videos available. To reach all sectors of the economy, we organized our dataset into 22 categories, totaling 60,000 videos. Using synthetic data, we generated a database of 700,000 videos that are used for training and testing. This technique allows us to reinforce the diversity of the data, as well as capture patterns and reduce biases.

Our Engagement Prediction model uses machine learning to estimate views, likes, and comments based on video metadata. As input variables, we extract and process the title and description, converting this information into numerical vectors (Embeddings), in addition to these variables we use duration of video feeding the model as well. We define the regression algorithm and optimize hyperparameters through supervised tuning. Performance evaluation is performed using three main metrics: RMSE (Root Mean Square Error), Explained Variance Score (EV-score), and Median Absolute Percentage Error (MdAPE). The training process is iterative, refining the parameters until an optimal point of accuracy is reached. After tuning, the model is applied to predict engagement metrics for new videos in a scalable and automated way.