Project 1: Understanding the Variables that Shape Flight Booking Choices Link
- Analyzed 50,000 data samples to uncover key factors influencing flight booking decisions.
- Employed data preprocessing and machine learning techniques to predict bookings and achieved a high 92% prediction accuracy using the Random Forest algorithm.
- By analysing variables that affect flight bookings we can focus more on those variables which are helping to get bookings.
- Such models can be used in many industries like e-commerce, digital marketing, Supply Chain and Logistics and many more.
- Made a Power BI dashboard to present insights.
Features That Affect Flight Bookings
![Screenshot 2023-08-17 191027](https://github.com/roshan9900/Roshan_Portfolio/assets/115538447/b728254a-4cb7-47a0-a6eb-ed79bd701f26)
- Utilized BERT for state-of-the-art text classification by fine-tuning the BERT Model to classify if the tweets are really related to disasters or not.
- Achieved top 25% ranking on Kaggle with an Overall accuracy of 85% with a weighted f1 score of 0.85.
- The dataset was too small to train such models, after a lot of combinations I was able to train the BERT with a very good accuracy.
- Such applications can be used to monitor social media posts and news headlines to check if the posts and headlines are real or not.
Model Architecture
![Screenshot 2023-08-17 191027](https://github.com/roshan9900/Roshan_Portfolio/assets/115538447/cb419bc8-02d8-4012-bf0e-16203e815f44)
Project 3: Text Generation WebApp Using Deep Learning Link
- Created a Text Generation Web App using Flask and Keras.
- Enabled users to input a starting phrase and receive AI-generated text.
- Demonstrated integration of AI and web development for an interactive experience.
![image](https://github.com/roshan9900/Roshan_Portfolio/assets/115538447/0e1b26bf-a77b-44de-b5a9-561f1fe37f32)
Project 4: Game Recommendation System Link
- Used Flask, Pandas, scikit-learn libraries to make a web application that recommends games to users and was hosted on the Google Cloud Platform.
- The dataset has over 200k rows and columns containing information about users.
- Converted playing hours into ratings to quantify user preferences and enhance accuracy.
- Such applications can be used in E-commerce, Streaming Services (Movies, Music, Videos), Social Media, News and Content Aggregation, Search Engines such industries.
![Screenshot 2023-08-17 191027](https://user-images.githubusercontent.com/115538447/261802808-5c054193-a531-4571-8c1b-37fc150d4ecb.png)