This project is based on the dataset from Fake job prediction kaggle dataset. In this project the task is to predict the fake job postings. There are 18 columns like title, department, company profile, job description etc. I will use NLP Techniques to analyse text data and prepare them for prediction model to train. Also this complete project uses pipeline concept automating the train and test process. Once the training is complete the model will be saved as a pickle file to load whenever we need to predict.
There are 17880 samples in the dataset in which 17014 are real job data and 866 are fake job data. The data is biased towards real job data so the main parameters used to rate prediction is not just accuracy but F1-score, Precision and recall as well. Before starting the whole coding process I have split the data into two datasets Train with 97% split and Test with 3% split. For the model I have used Random Forest Classifier which at the end has achieved F1-score 0.80 and Precision of 94.4% with accuracy around 98%.
This is a overall view of the files and functionalities used in both the pipelines in the project. Below I have attached both train and test pipelines separately.
The training process contains 3 major steps
- Data Reading
- Data Handling
- Model
First we read the data using Data Read file and then pass it to Data handling file where the missing values, Text data Handling takes place and finally categorical encoding is done for category columns. After Handling the data and scaling then we will utilize Random Forest algorithm to create and train model on the cleaned data. After training the model then we save the model as a pickle file so it can be loaded any time.
Similar to training we need to read the data using Data Read file and then handle the data(missing values, Text data Handling, categorical encoding). After Handling the data and scaling them we will make use of the saved model to test the data and save the predicted result as a csv. Here for testing process we use the saved categorical encoder and saved scaler instead of newly creating. Also in the end it will display the metrics used for rating.
For columns like title, department, company profile, job description etc for all the categorical I have used NLTK functions like stop words, Stemming and Word net synset to find out a similarity factor between columns eg: similarity between job title and job description. Using various combinations I have formed various new columns which represent such kind of similarities. Now all these columns are used in the Random Forest algorithm to find the Fake jobs. With all these steps I have reached precision of 94% and f1-score of 0.80(Max f1-score is 1 so its like 80%).
After downloading the files to your local computer just open terminal in the folder where Main.py file is present and give below commands.
By default the code takes training as the process to perform. You can also specify with -r parameter as in below picture.
Just specify -r test so that the testing pipeline starts. If it is not specified then default training pipeline will initiate.