Naïve Bayes and SGD

Using Naive Bayes and SGD ML algorithms to predict if a new car for sale will decrease in price.

The Data Set

Example:

Vin	Make	Model	Trim	Color	Dealership	Location	Price Change*
1X234...	Volkswagen	Jetta	SE	Red	Brady	VW of Boston	1000
4C353...	BMW	x5	XT	Green	Elway	BMW of Denver	0
9E434...	Volvo	s60	SV-600	Blue	Rodgers	Volvo of GB	52
...	...	...	...	...	...	...	...

*price change must be > $100 to be considered to have changed

Data Set Size

Total	13311 cars
# price change	6067 cars
# no price change	7244 cars

Procedures

Corpus

The data set was split into a traning set and test set. For each row in the database, each column was concatenated to form a list of strings called the corpus.

Example:

Training corpus: ["volkswagen jetta se red brady vw of boston", "bmw x5 xt green elway bmw of denver"]

Test corpus: ["volvo s60 sv-600 blue rogers volvo of gb"]

Targets

The target vector indicates which class each training example is in. 0 represents no change in price while a 1 represents a change in price.

Example:

Training Target: [1,0]

Test Target: [0]

Results

	precision	recall	f1-score	support
no_change	0.88	0.80	0.84	2197
change	0.79	0.87	0.83	1866
avg / total	0.84	0.83	0.83	4063

Overall acurracy on test set: 83.2%