Implementation of a Naïve Bayesian classifier that calculates and reports the probability that 10 unidentified objects belongs to one of two classes: airplanes and birds.
CS 131 HW 05 - Naive Bayesian Classification
Brandon Dionisio
Using a Naïve Recursive Bayesian classifier, calculate and report the probability that 10 unidentified objects belongs to one of two classes.
The likelihood distribution for airplanes is in the first row of likelihood.txt and the likelihood distribution for birds is in the second row of likelihood.txt
Ten tracks representing the velocity of the unidentified flying object measured by a military-grade radar (1s sampling frequency for a total length of 600s) in training.txt
If the radar could not acquire the target and perform the measurement, the corresponding data point is a NaN. These tracks are raw data.
Twenty tracks representing the velocity of the birds and airplanes (10 rows of birds followed by 10 rows of airplanes in training.txt) measured by a military-grade radar (1s sampling frequency for a total length of 600s). If the radar could not acquire the target and perform the measurement, the corresponding data point is an NaN value. These tracks are curated to have a maximum sample drop rate of 5% of the total number of samples per track.
stackoverflow
CS 131 Canvas Slides
scikit-learn.org
To run the radar, use "python radar.py"
None
Feature 1: Likelihood distribution of the class based on speed as given by likelihood.txt
Feature 2: Likelihood distribution of the class based on variance. This distribution was divided up as the variance of speeds every 6 seconds.
To preprocess the training data and the testing data, all NaN values are turned into the mean for each row.
To prevent 0 probabilities in the likelihood distributions, I added 0.001 to each probability. This would prevent any invalid predictions or divisions by 0.
In normalizing the data, I divided the combined probabilities for both classes by 2.4 as this is the total probability from the addition of 0.001 for both features.