Predicting OTT Subscription Behavior Using Deep Learning: A Multi-Class Classification Approach with Feature Engineering and Class Imbalance Handling

With the introduction of different OTT streaming platforms like Netflix, Amazon Prime Video, and Disney+ Hotstar, subscription user behavior has become important factors in building customer retention, pricing, and content recommendations. These methodologies do not address the non-linear relationships that exist between important variables like pricing, quality of content, discount availability, and payment options. This study forms the basis for developing a deep learning approach to predicting subscription behavior for OTT platforms based on a multi-layer perceptron (MLP) model. Preprocessing of the survey data for feature selection, label encoding, and normalization will include addressing the class imbalance on the dataset using SMOTE for balanced learning. The architecture contains batch normalization and dropout regularization, with training carried out using Adam optimizer for 500 epochs. Model application achieved a very high accuracy of $97 \%$, which is further validated on various classification metrics as well as confusion and accuracy-loss plots. The study pinpoints pricing, promotions, and satisfaction with content as the prominent factors responsible for subscription choices.