The massive popularity of social media platforms and the extensive engagement of people on these platforms have made it easier to express opinions or sentiments through posts, comments, and messages, generating a significant amount of digital text data every day. Sentiment analysis is the process of summarizing and categorizing opinions expressed on social media. Detecting sentiment in the Bangla language has become increasingly important due to a notable lack of sentiment detection capabilities for informal Bangla text. This research aims to develop an advanced sentiment detection model for analyzing sentiment in informal, noisy Bangla text collected from social media comments. Our major contribution involves using web scraping to gather comments from different sites, resulting in a large dataset of more than 3000 comments. This approach minimizes bias compared to single-source data and captures sentiment across a wide range of topics. To ensure accuracy and dependability, we collaborated with psychology students to carefully classify the gathered comments by sentiment polarity. We built a customized deep-learning strategy using LSTM and optimized several LSTM model hyper parameters, including learning rate, batch size, number of epochs, dropout rate, and number of LSTM layers. Our customized deep-learning strategy proved beneficial, as the LSTM model outperformed other models on informal Bangla text. The research achieved 80.3% accuracy for the Bangla informal dataset and 96.9% accuracy for the formally written Bangla text dataset, underscoring the efficiency of the proposed approach. Finally, a user-friendly interface was built to capture comments from social media and classify their sentiments.