Fraudulent Job Posts

NLP: K Nearest Neighbors

What Is k-NN? In pattern recognition, the k-nearest neighbors algorithm (k-NN) is a non-parametric method used for classification and regression. In both cases, the input consists of the k closest training examples in the feature space. [Wikipedia]

Classification Report: k-NN

With Parameter Tuning

	precision	recall	f1-score	support
Real	0.91	0.95	0.93	754
Fake	0.76	0.63	0.69	194
Accuracy			0.89	948
Macro Avg	0.84	0.79	0.81	948
Weighted Avg	0.88	0.89	0.88	948

Classification Report: k-NN

Without Parameter Tuning

	precision	recall	f1-score	support
Real	0.88	0.95	0.92	754
Fake	0.73	0.51	0.60	194
Accuracy			0.86	948
Macro Avg	0.81	0.73	0.76	948
Weighted Avg	0.85	0.86	0.85	948

Train Time: 0.002 seconds

Findings: The K Nearest Neighbors model was by far the most efficient, but not very reliable in terms of accuracy. It took .0025 seconds, exponentially less time than the other models. However, before the hypertuning, the model had a precision score of .73 with regard to fraudulent postings and an overall accuracy score of .86, not very impressive. After using gridsearchCV to calculate the best parameters, the model’s accuracy increased to .89 and it’s precision score increased to .76, but this is still quite low. I would not be confident in this model’s ability to correctly identify a fake job posting.

Sample Plot: k-NN NLP

Model Comparison

	Model	Precision Fake
NLP	k-NN	0.76
	SVM	0.89
	DL	0.92
	NB	0.84
	RF	0.58
Non-Textual	k-NN	0.85
	SVM	0.84
	DL	0.86
	RF	1.00