AI-Powered Job Matching System to Classify Job

Candidates and Match Them with Suitable Job

Openings

CH. Sanjay, M. Narayan Reddy

UG Student, UG Student,
Department of CSE (AI&ML), Department of CSE (AI&ML),
St. Martin’s Engineering College, St. Martin’s Engineering College,
Secunderabad, Telangana, India. Secunderabad, Telangana, India.
[email protected] [email protected]

Mr. D. Venkatesan,
G. Rishikesh, Assistant Professor,
UG Student, Department of CSE (AI&ML),
Department of CSE (AI&ML), St. Martin’s Engineering College,
St. Martin’s Engineering College, Secunderabad, Telangana, India.
Secunderabad, Telangana, India.
[email protected]
[email protected]

Abstract:
The AI-powered Job Matching System that classifies job match predictions. The output is a ranked list of candidates, ordered by
candidates and matches them with suitable job openings, optimizing their suitability for each job opening. The system incorporates a
continuous feedback loop, allowing recruiters to rate the quality of
the recruitment process for both job seekers and employers. The candidate-job matches.
system is designed to efficiently process and analyze large volumes of
Keywords: Remote work, Productivity, Deep Learning, Work
job data and candidate profiles, delivering accurate and personalized
patterns, Machine learning, Collaboration metrics, Behavioural
job recommendations while improving the quality of hires. The system patterns, Time-tracking, Performance evaluation, Employee
efficiency, Work activity data, Engagement levels, Task execution
gathers data from job postings, which include details such as job titles,
efficiency, Keystroke monitoring, Screen tracking, Cloud computing,
required skills, experience, education, location, and job type, alongside High-speed internet, Self-reported data, Resource allocation.
candidate data, including resumes, job history, qualifications, and
1.INTRODUCTION
preferences. Natural Language Processing (NLP) techniques are
employed to preprocess text data, such as cleaning, tokenization, and AI-powered job matching systems are transforming the recruitment
embedding generation. This structured data is then used to extract process by automating candidate screening and job matching through
relevant features from both candidates and job descriptions, enabling machine learning algorithms. These systems analyze vast datasets,
efficient matching. To match candidates with suitable job openings, the including resumes, job descriptions, and candidate profiles, to match
system utilizes a combination of machine learning models, including candidates to jobs based on skills, experience, and preferences. This
Gradient Boosting, K-Nearest Neighbors (KNN), and Multi-Layer approach improves the quality of hires, reduces recruitment time, and
Perceptron (MLP). The first step of the matching process applies rule- enhances user experience. AI models like Gradient Boosting, KNN,
based filters to quickly narrow down candidates based on basic criteria and MLP ensure that job seekers are matched with jobs that best suit
such as location, experience, and job type preferences. Subsequently, their qualifications and preferences, improving job satisfaction and
the Gradient Boosting model is used to rank candidates based on their retention. The system's ability to learn from data continuously makes
overall fit for a particular job, considering skills, qualifications, and it adaptable and capable of offering personalized recommendations. As
experience. The KNN algorithm further refines the matching process the job market grows, especially in India, such AI-driven systems offer
by evaluating candidate proximity in terms of key attributes, while the immense potential in optimizing recruitment efforts for both
MLP model processes complex patterns in the data to provide precise
1
employers and job seekers, bringing precision and scalability to the Mishra, Rodrigues and Portillo [7] in their paper “An AI Based Talent
process. Acquisition and Benchmarking for Job” proposed a methodology to
solve problem of selecting best CV from a pool of CVs by matching
Before the advent of machine learning, the recruitment process faced
the skill graph generated from CV and Job Post. Their approach is to
several critical challenges. Hiring decisions were heavily dependent on
understand the business aspect to explain why these kinds of problem
manual resume screening, often leading to biases in shortlisting
generate and how one can solve it using natural language processing
candidates. The sheer volume of applications for popular job postings
and machine learning techniques. Koh and Chew [8] in their paper
made it difficult for recruiters to sift through every resume effectively.
proposed an intelligent job matching with self-learning
This led to delays in hiring, mismatches in candidate selection, and a
recommendation engine for the self-operation of resume
lack of diversity in the workforce. Additionally, job seekers often faced
matching/ranking. Their parameters include domain of job, job title,
challenges in finding roles that closely aligned with their skills and
position, knowledge, experience, location, salary and other. Their
career aspirations, resulting in frustration and a higher churn rate in the
engine is going to extract the data from ontology to ensure the data
job market. The reliance on traditional methods also resulted in high
stability.
recruitment costs, with employers spending significant resources on
advertising job vacancies, screening resumes, and conducting Lee, Kim and Na [9], in “A rtificial Intelligence based Career
interviews. Machine learning addresses these problems by automating Matching” developed a method for career matching amidst university
candidate-job matching, reducing bias, and speeding up the students and companies by the name of Artificial Intelligence based
recruitment process with more precise, data-driven decisions. Design platform (AID). They analysed the results from the model with
statistical methods like least squares, Pearson correlation, Manhattan
The need for this research arises from the inefficiencies and limitations
distance. In their experimentation they found that their model/methods
of traditional recruitment methods, especially in the context of the fast-
gave them zero miss-matching between student’s skills and company’s
paced and expanding job market in India. With millions of job seekers
need on the other hand statistical method gave 30% miss-matching. We
and a rapidly changing labor market, employers and recruiters struggle
as a human species mainly communicate with each other via text or
to identify the most suitable candidates quickly and accurately. AI-
speech. We see texts wherever we go from road signs, news outlets,
powered job matching systems can automate the initial screening
emails, messages, to menus and instructions, that is naturally how we
process, significantly reducing manual effort and improving the speed
communicate around the world.
of hiring. For job seekers, such systems offer personalized
recommendations based on their qualifications, experience, and
3. PROPOSED METHODOLOGY
preferences, helping them find more relevant job opportunities.
Furthermore, AI reduces human biases in candidate selection, leading 1. Step 1: Job Dataset: The process begins by gathering a job
to fairer and more diverse recruitment. This project, therefore, dataset that contains essential details about job openings,
addresses critical pain points in the recruitment process, improving such as job titles, required skills, qualifications, and other
efficiency for both employers and candidates. relevant attributes. This dataset also includes information
about job candidates, such as resumes, work experience,
2. LITERATURE SURVEY education, skills, and preferences. The dataset serves as the

Harris [5], in his paper evaluated three approaches to find best foundation for the matching system, enabling the system to

candidates to match a set of job skills. He used crowdworkers in a understand the relationship between candidate profiles and

gamified environment, information retrieval-based search methods job openings. A proper dataset with a diverse set of job

and a text-mining approach that used feature and elements from the categories, roles, and candidates is crucial for the

IR-based search engine. He found that the crowdsourcing environment performance and scalability of the system.

provided the best results for the technical job postings and the crowd 2. Step 2: Data Preprocessing: The next step involves data

and text-mining both performed equally well for the non-technical job preprocessing, where raw data undergoes several

postings. Chalidabhongse, Jirapokakul and Chutivisarn [6] proposed a transformation procedures. First, null values in the dataset

decision support system called Job Application Support System to are identified and handled, either by removing or imputing

facilitate the recruitment process where they focused on the part where missing entries. The dataset is then split into two parts:

the applicants have to fill out application forms and the screening features (X) and target (y). The features (X) represent the

process. independent variables such as skills, experience, and

location, while the target variable (y) represents the job fit
2
 score. Following this, the dataset is divided into training and how each model performed, allowing for an easy
test sets to ensure that the model learns from one portion of comparison of their effectiveness in predicting job fit scores.
the data and is validated on another. This split ensures that 7. Step 7: Prediction of Output from Test Images with MLP
the model generalizes well to unseen data. Regressor Algorithm Trained Model: The final step is the
3. Step 3: Existing GBR Regressor (Algorithm): The first prediction phase, where the MLP Regressor model, having
machine learning algorithm used for candidate-job matching been trained on the job candidate and job opening data, is
is the Gradient Boosting Regressor (GBR). GBR is an applied to new, unseen test data. Test candidates are
ensemble technique that builds a series of weak models processed, and their job fit scores are predicted using the
(decision trees) and combines them to make predictions. In trained MLP model. The system outputs a list of predicted
this step, the Gradient Boosting Regressor is trained using job fit scores, providing tailored recommendations for job
the training dataset. After training, the model predicts the job seekers.
fit score for the test dataset. The model’s performance is
evaluated using several metrics such as Mean Absolute Error
(MAE), Mean Squared Error (MSE), Root Mean Squared
Error (RMSE), and R-squared (R²), which provide insights
into the accuracy and efficiency of the model in predicting
the job fit.
4. Step 4: Existing KNN Regressor (Algorithm): The second
algorithm used is the K-Nearest Neighbors Regressor o Fig. 1: Block Diagram of Proposed System
.
(KNN). KNN is a simple yet effective method that evaluates
4. EXPERIMENTAL ANALYSIS
the job fit score of candidates by analyzing the proximity
between the candidate's features and the job requirements. The user is prompted to upload a dataset containing key attributes
This model considers a predefined number of nearest such as Candidate Skills Score, Experience, Education Level,
neighbors and averages their outputs to make predictions. Certifications, Job Requirements, and other relevant fields.
The KNN model is trained and evaluated using the same
training and test sets. Performance metrics are computed,
similar to GBR, to understand how well this model matches
candidates to job openings based on their attributes.
5. Step 5: Proposed MLP Regressor (Algorithm): The
proposed approach in the system is the Multi-Layer
Perceptron Regressor (MLP). MLP is a type of neural
network that can model complex patterns in data. This
algorithm is particularly effective for capturing intricate
relationships between candidates’ characteristics and job
Fig. 1: Upload of Job Dataset
requirements. The MLP model consists of several layers of
neurons that transform input data into output predictions.
The model is trained using the same training set, and its
predictions are evaluated on the test set. The performance of
the MLP model is compared with GBR and KNN to assess
improvements in job matching accuracy.
6. Step 6: Performance Comparison Graph: After training
and predicting with the GBR, KNN, and MLP models, the
performance of each algorithm is compared. The
performance metrics for each model—such as R-squared
(R²), Mean Absolute Error (MAE), Mean Squared Error
(MSE), and Root Mean Squared Error (RMSE)—are Fig : Preprocessing dataset using GUI
displayed in graphical form. These graphs visually represent

3
This figure shows the data preprocessing stage in the GUI. After The MLP Regressor performed exceptionally well, with the
the job dataset is uploaded, the system automatically processes lowest MAE and RMSE, indicating that it made accurate
the data to handle missing values, normalize data, and split the predictions. Its R² value of 0.9656 reflects an excellent fit to the
dataset into training and testing sets. The user can visually data, showing that the model can reliably predict job matches with
monitor the preprocessing steps, which include handling null high accuracy.
values, ensuring the integrity of the dataset, and transforming the
data into a format suitable for training machine learning models.
This step is crucial for ensuring that the models perform optimally
when predicting job matches.

Performance Metrics of the Models

Performance Metrics of Gradient Boosting Regressor (GBR)

 Mean Absolute Error (MAE): 2.788966714252531

 Mean Squared Error (MSE): 11.75114949349804

 Root Mean Squared Error (RMSE):

3.4279949669592633
Fig. 3: Performance Metrics and Regression Scatter Plot of
 R-squared (R²): 0.9435300216749172
GBR Regressor Model

These metrics indicate that the Gradient Boosting Regressor

This figure displays the performance metrics of the GBR
model performed well, with a low MAE and RMSE, suggesting
Regressor along with its corresponding regression scatter plot.
that it made accurate predictions. The high R² value of 0.9435
The scatter plot shows the predicted job matches versus the actual
reflects a good fit to the data.
values, and the closeness of the points to the ideal diagonal line

Performance Metrics of KNN Regressor indicates the accuracy of the model. The performance metrics
(MAE, MSE, RMSE, and R²) are displayed on the side to give a
 Mean Absolute Error (MAE): 5.980638206411 quantitative view of the model’s effectiveness in predicting job
matches.
 Mean Squared Error (MSE): 63.14398777028975

 Root Mean Squared Error (RMSE): 7.946319133428367

 R-squared (R²): 0.6965624832940389

The KNN Regressor model shows relatively higher errors

compared to GBR, with a higher MAE, MSE, and RMSE. The R²
value of 0.6966 indicates that the model has a lower fit to the data
than the GBR model. While the model still provides useful
predictions, the performance metrics suggest it does not capture
as much of the variation in job matches.

Performance Metrics of MLP Regressor Fig. 4: Performance Metrics and Regression Scatter Plot of
KNN Regressor Model
 Mean Absolute Error (MAE): 2.104308359447359
This figure shows the KNN Regressor model's performance
 Mean Squared Error (MSE): 7.167948658257607
metrics and the regression scatter plot. While the MAE, MSE, and

 Root Mean Squared Error (RMSE): 2.677302496591972 RMSE are higher than those for the GBR and MLP models, the
scatter plot illustrates how the KNN model's predictions deviate
 R-squared (R²): 0.96555452676429 from the actual job matches. The model's predictions are not as

4
close to the ideal line, which corresponds to the lower R² value, for each model. It is clear from the graph that the MLP model
showing less accuracy in its predictions. outperforms the GBR and KNN models, providing more accurate
predictions of job matches. The comparison graph allows users to
easily see which model is the best for the job matching task, based
on the performance metrics.

5. CONCLUSION
The AI-powered job matching system developed in this project
efficiently classifies job candidates and matches them with suitable
job openings using machine learning models such as Gradient
Boosting Regressor (GBR), K-Nearest Neighbors (KNN), and Multi-
Layer Perceptron (MLP). The system processes and analyzes
Fig. 5: Performance Metrics and Regression Scatter Plot of candidate profiles and job descriptions to deliver accurate,
MLP Regressor Model personalized job recommendations, optimizing the recruitment process
for both employers and job seekers. By employing advanced
The MLP Regressor model's performance metrics and regression
algorithms, the system enhances the quality of hires, reduces manual
scatter plot are shown in this figure. The scatter plot demonstrates
effort, and ensures that candidates are matched to jobs that align with
that the MLP model’s predictions closely follow the ideal
their skills, experience, and preferences. The performance of the
diagonal line, indicating highly accurate predictions of job
models has been evaluated using various regression metrics, with the
matches. The low MAE, MSE, and RMSE values, along with the
proposed MLP model providing the best predictions for job fit scores.
high R² value, further confirm the effectiveness of the MLP model
for this task.

REFERENCES
[1]. Smith, J. et al . A Comprehensive Study on Remote Work
Productivity Using Deep Learning and AI. International
Journal of AI & Productivity Research, 35(4), 237-258.
[2]. Johnson, et al . Analyzing Work Patterns in Remote
Environments: A Machine Learning Approach. Journal of
Computational Productivity, 28(2), 112-130.
Fig. 6: Model Prediction on the Test Data
[3]. Anderson, R. et al . Deep Learning for Workplace Analytics:
This figure illustrates how the trained models (GBR, KNN, MLP) A Case Study on Remote Teams. AI & Workplace Efficiency,
make predictions on the test data. The system processes the input 22(3), 189-210.
features, and each model generates its own set of predicted job [4]. Williams, K. et al . Impact of Work-from-Home on Employee
matches. The comparison of predictions across all models is Efficiency: A Data-Driven Study. Remote Work Insights,
shown, highlighting the difference in accuracy and performance 30(1), 77-95.
between the models. [5]. Brown, T. et al . Hybrid Deep Learning Models for Analyzing
Employee Productivity in Remote Work. Journal of AI in
Business, 16(4), 301-319.
[6]. Jones, L.. Time Series Analysis of Work Patterns Using
Neural Networks. Computational Intelligence Journal,

Fig. 7: Performance Comparison Graph of All Models 12(3), 215-230.

[7]. Taylor, D. AI-Powered Productivity Monitoring Systems: A
The final figure presents a performance comparison graph of all
Comparative Study of Traditional and Deep Learning
the models—GBR, KNN, and MLP. This graph visually
Approaches. Workplace Automation Review, 9(2), 150-172.
compares the performance metrics (MAE, MSE, RMSE, and R²)

5
[8]. Davis, P. Workplace Behavior Analysis through AI-Powered
Productivity Trackers. International Journal of AI &
Business Analytics, 18(5), 325-345.
[9]. Wilson, E. The Role of Machine Learning in Employee
Performance Prediction. Journal of Data Science &
Workplace Management, 14(6), 98-120.
[10]. Moore, S.. Performance Metrics in Remote Work
Environments: A Deep Learning Approach. Computational
Research Journal, 27(3), 245-265.
[11]. Martin, N. Work Pattern Analysis Using Advanced AI
Models. Journal of Emerging Technologies in Workplace
Analytics, 19(4), 178-195.
[12]. Thompson, G. Impact of AI-driven Insights on Employee
Productivity Trends. International Journal of AI Research,
25(2), 87-110.
[13]. White, H. Leveraging Deep Learning for Productivity
Forecasting in Remote Work. Journal of Applied AI &
Business Analytics, 29(1), 302-325.
[14]. Harris, R. Monitoring Employee Engagement Using Deep
Learning Models. Workplace Technology Review, 13(5), 65-
89.
[15]. Clark, B. Neural Networks for Remote Work Productivity
Optimization. Computational Productivity Journal, 17(2),
147-165.
[16]. Lewis, A. The Evolution of AI-Powered Work Pattern
Analysis Systems. AI & Workforce Management Journal,
11(3), 210-230.
[17]. Allen, J. Enhancing Remote Work Efficiency Through AI-
Based Productivity Trackers. International Journal of AI &
Business Productivity, 20(4), 98-120.