The current TSA body scanner passenger screening algorithm has a high rate of false positives, which leads to longer wait times at screening checkpoints. Using the TSA Kaggle dataset, I developed models to predict potential threats in two body zones of 3D body scan images. The current best model has 95% accuracy, 67% precision, and 86% recall. This is the first project I've done in Python, and am happy to say I've met many of my goals for the ChiPy mentorship program!

1. Building a Better TSA Screening Algorithm:
a ChiPy Mentorship Project
Yvonne K. Matos
ChiPy Meeting, December 14, 2017
Photo: Benoit Tessier/Reuters

2. My Goals, ChiPy Mentorship Program
1. Get more experience with Python
2. Develop deep learning algorithms
3. Work with large datasets and cloud computing
4. Get hired as a data scientist!
Data Science Pipeline
Data science is 80% cleaning and preprocessing, 20% modeling
Business
Question
Data
Question
Data
Collection
Data
Loading
Data
Cleaning
PreprocessingModelingValidation
Data
Answer
Business
Answer
Exploratory
Analysis

3. TSA Passenger Algorithm Screening Challenge
Problem: High false alarm rates create bottlenecks at airport checkpoints.
Challenge: Create an algorithm with a lower false alarm rate using a dataset of scan
images with simulated threats
Business
Question
Data
Question
Data
Collection
Data
Loading
Data
Cleaning
PreprocessingModelingValidation
Data
Answer
Business
Answer
Exploratory
Analysis

4. Anticipated Challenges
• First Python project
• Huge files = long run times
• Enormous scope
– Full training in cloud = $$$$$
Plan of attack
• Begin small
• Scale up locally
• Run in the cloud

5. Exploration: Visualizing the Images
(n, 512, 660, 16)
Raw data for lowest res 10 MB image: 4D array
• 3D images
• 1,147 total
• 3TB dataset
(n, 512, 512, 660)
Higher res 330 MB image:
Samples
Dimensions
Channels

6. Start with two body zones
17 body zones * 1,147 images = 19,499 potential threats
Zone 17
• 1052 non-threat
• 95 threat (8.3%)
Zone 5
• 1041 non-threat
• 106 threat (9.2%)

7. Preprocessing:
3D to 2D Classification Problem
16 channels
(n, 512, 660, 16)
Raw data 10 MB image: 4D array
Zone 5 cropped
Zone 17 cropped
Crop images and save to PNG format

8. Developing Models: Convolutional Neural Network
classifier = Sequential()
classifier.add(Convolution2D(32, (3, 3), input_shape=(250, 250, 3), activation='relu'))
classifier.add(MaxPooling2D(pool_size=(2, 2)))
classifier.add(Convolution2D(32, (3, 3), activation='relu'))
classifier.add(MaxPooling2D(pool_size=(2, 2)))
classifier.add(Convolution2D(64, (3, 3), activation='relu'))
classifier.add(MaxPooling2D(pool_size=(2, 2)))
classifier.add(Flatten())
classifier.add(Dense(units=128, activation='relu'))
classifier.add(Dense(units=256, activation='relu'))
classifier.add(Dense(units=1, activation='sigmoid'))
classifier.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])
from keras.preprocessing.image import ImageDataGenerator
train_datagen = ImageDataGenerator(rescale=1./255, shear_range=0.2,
zoom_range=0.2, horizontal_flip=True)
test_datagen = ImageDataGenerator(rescale=1./255)
training_set = train_datagen.flow_from_directory('C:PathZ17_training_png',
target_size=(250, 250), batch_size=32, class_mode='binary')
test_set = test_datagen.flow_from_directory('C:PathZ17_test_png',
target_size=(250, 250), batch_size=32, class_mode='binary')
classifier.fit_generator(training_set, steps_per_epoch=917, epochs=10,
validation_data=test_set, validation_steps=230)
classifier.save('C:PathZ17_classifier.h5')
Convolutional Layers
Fully Connected Layers
Image Augmentation

9. Preliminary Pipeline Zone 17 Model
Accuracy: 95%
Precision: 67%
Recall: 86%
Zone 5 Model
Accuracy: 78%
Precision: 27%
Recall: 77%
1 row = 1 body scan image
Body scan
image
Crop, PNG Model
Prediction
Prediction
If
Pred. threat
FLAG!!

10. Thanks!
• Thanks for coming tonight
• ChiPy mentorship program
• mHUB for hosting