1
PROJECT PRESENTATION FLIGHT DELAY PREDICTION EXPLORATORY DATA ANALYSIS FEATURES RESULTS ANALYSIS RANDOM FOREST FEATURE ENGINEERING FEATURE ANALYSIS OPTIMIZATION OF THE RANDOM FOREST GAUSSIAN NAIVE BAYES - Model: - Training: - Model: - Training: DATASETS: our main dataset includes every domestic flight from 2001 to 2008, with scheduled times, actual times, airpo- rts (depature/arrival), delay (if any) and cause of delay, plane identifier LOGISTIC REGRESSION NEURAL NETWORK ALGORITHMS GOAL : to predict if a flight will be delayed at arrival or not CHALLENGES : - hard to train a model to capture the delay of previous flights - imbalanced dataset between delayed and non delayed exa- mples NEXT STEPS: - Improve the Neural Network (better shape) - Increase size of the sample of training examples with data from other years - Add new features to capture non linearity (we still have no overfitting) - Weights for the logistic re- gression on normalized fea- tures -Several features of force memory are present F1-score vs max_depth F1-score vs min_samples

FLIGHT DELAY PREDICTION - Machine Learningcs229.stanford.edu/proj2016/poster/... · Terrorist attacks around the world with US fatalities Type of plane and characteristics (seats,

  • Upload
    others

  • View
    1

  • Download
    0

Embed Size (px)

Citation preview

Page 1: FLIGHT DELAY PREDICTION - Machine Learningcs229.stanford.edu/proj2016/poster/... · Terrorist attacks around the world with US fatalities Type of plane and characteristics (seats,

PROJECT PRESENTATION

FLIGHT DELAY PREDICTION

EXPLORATORY DATA ANALYSIS

FEATURES RESULTS

ANALYSISRANDOM FOREST

FEATURE ENGINEERING

FEATURE ANALYSIS OPTIMIZATION OF THE RANDOM FOREST

GAUSSIAN NAIVE BAYES

- Model:

- Training:

- Model:

- Training:

DATASETS: our main dataset includes every domestic flight from 2001 to 2008, with scheduled times, actual times, airpo-rts (depature/arrival), delay (if any) and cause of delay, plane identifier

LOGISTIC REGRESSION

NEURAL NETWORK

ALGORITHMS

GOAL : to predict if a flight will be delayed at arrival or notCHALLENGES : - hard to train a model to capture the delay of previous flights- imbalanced dataset between delayed and non delayed exa-mples

NEXT STEPS: - Improve the Neural Network (better shape)- Increase size of the sample of training examples with data from other years- Add new features to capture non linearity (we still have no overfitting)

- Weights for the logistic re-gression on normalized fea-tures-Several features of force memory are present

F1-score vs max_depth F1-score vs min_samples