Home | Portfolio | Terms and Conditions | E-mail me | LinkedIn
Predicting Hotel Cancellations with Machine Learning
- Employed pandas to process, sort, and aggregate over 79,000 booking entries by total cancellations per week.
- Implemented machine learning workflow using Amazon SageMaker – used Git to commit relevant repository.
- Used scikit-learn to implement ExtraTreesClassifier for feature selection and SVM for classification. Identified lead time, deposit type and country of origin as cancellation drivers. Achieved AUC of 0.74 on the test set.
- Devised LSTM model with TensorFlow 2.0 to predict weekly cancellations. Model showed improvement in RMSE from 57.95 (SARIMA performance) to 31.98 (LSTM performance).
- Part 1: Support Vector Machines and ARIMA
- Part 2: Classification with Keras Neural Network
- Part 3: Time Series Predictions with LSTM Network
Image Recognition with Keras: Convolutional Neural Networks
- Built classifier to distinguish between images of cars and planes. Used training and validation data of 200 images.
- Implemented data augmentation with VGG16 pretrained network to prevent overfitting and trained model across 30 epochs. Validation loss significantly reduced from 0.9405 to 0.2567.
- Model correctly categorized 93% of test images.