![Deepfake Image Detection App](http://www.jeffspagnola.com/wp-content/uploads/2020/12/Screen-Shot-2020-12-15-at-8.39.05-PM-1-1024x699.png)
Deepfake Image Detection App
![Deepfake Image Detection Notebook](http://www.jeffspagnola.com/wp-content/uploads/2020/12/deepfake-video-1024x576.jpg)
Deepfake Image Detection Notebook
![Detecting Disaster Tweets](http://www.jeffspagnola.com/wp-content/uploads/2020/12/dark-twitter-birdjpg-1024x768.jpg)
Detecting Disaster Tweets
![Global Terrorism Analysis](http://www.jeffspagnola.com/wp-content/uploads/2020/12/main-image-1024x514.jpeg)
Global Terrorism Analysis
-
Deepfake Image Detection App
Deployed app that is associated with the Deepfake Image Detection Notebook. This app was constructed using Flask along with manually built html pages. The app allows a user to input an image and uses a finely tuned CNN to predict whether the image is real or a deepfake.
Note: As of this moment, this app is designed to work with high quality deepfake images. Screengrabs or poorly photoshopped images don't play well with the model. -
Deepfake Image Detection Notebook
Built a system that can predict whether an image was real or a deepfake with 97% accuracy and deployed the model to an associated app.
- Obtained 140k+ images, both real and deepfakes, by combining multiple datasets of human face images
- Processed the images into standard size and converted to arrays for modeling using ImageDataGenerator
- Developed several high accuracy CNN models to predict between real and deepfake images achieving a 97% accuracy
- Deployed a working app where users can test the model with a single image -
Detecting Disaster Tweets
Used NLP & Deep Learning to analyze disaster related tweets to make a social media monitor that determines whether a tweet is in reference to a disaster.
- Utilized NLP & visualizations to analyze a dataset containing over 11,000 tweets associated with disaster keywords
- Gained insight into the context and sentiment of tweets in the dataset by using NLP to find trends of disaster-related tweets
- Developed several types of machine learning models resulting in an 89% accurate Stacking Classifier
- Implemented several LSTM Networks for classification resulting in 90% accurate predictions -
Global Terrorism Analysis
Analyzed terrorist attack data and used various machine learning models to determine the factors that make an attack successful in order to help evaluate current security protocols.
- Obtained, scrubbed, and analyzed data on 181,000 terrorist attacks between 1997 and 2017
- Used various advanced feature engineering methods to prepare the data for modeling including KNNImputer & condensing columns
- Created several exploratory visuals to understand trends of successful attacks for the purpose of determining prevention methods
- Developed a series of machine learning models for classification resulting in a 96% accurate Stacking Classifier