Internship Report

(Project Work)
On

TRAINING PROGRAM RECOMMENDATION

SYSTEM
Submitted to
JAWAHARLAL NEHRU TECHNOLOGICAL UNIVERSITY ANANTAPUR, ANANTHAPURAMU
In Partial Fulfillment of the Requirements for the Award of the Degree of
BACHELOR OF TECHNOLOGY
In
COMPUTER SCIENCE & ENGINEERING (DATA SCIENCE)

Submitted By
VUSA MAHENDRA - (21691A3252)

MADANAPALLE INSTITUTE OF TECHNOLGY & SCIENCE

(UGC – AUTONOMOUS)
(Affiliated to JNTUA, Ananthapuramu)
Accredited by NBA, Approved by AICTE, New Delhi)
AN ISO 21001:2015 Certified Institution
P. B. No: 14, Angallu, Madanapalle, Annamayya – 517325

1
 MADANAPALLE INSTITUTE OF TECHNOLOGY & SCIENCE
(UGC-AUTONOMOUS INSTITUTION)
Affiliated to JNTUA, Ananthapuramu & Approved by AICTE, New Delhi
NAAC Accredited with A+ Grade, NIRF India Rankings 2022 - Band: 251-300 (Engg.)
NBA Accredited - B.Tech. (CIVIL, CSE, CST, ECE, EEE, MECH), MBA & MCA
www.mits.ac.in

DEPARTMENT OF COMPUTER SCIENCE & ENGINEERING (DATA SCIENCE)

BONAFIDE CERTIFICATE

This is to certify that the 20CSD702-Project Work & Internship entitled “TRAINING
PROGRAM RECOMMENDATION SYSTEM ” is a bonafide work carried out by

VUSA MAHENDRA - (21691A3252)

Submitted in partial fulfillment of the requirements for the award of degree Bachelor of
Technology in the stream of Computer Science & Engineering (Data Science) in
Madanapalle Institute of Technology & Science, Madanapalle, affiliated to Jawaharlal
Nehru Technological University Anantapur, Ananthapuramu during the academic year
2024-2025.

Head of the Department Internship Coordinator/CSD

Dr. S. Kusuma Mrs. Manjula Prabakaran

Assistant Professor & Head Assistant Professor
Department of CSE (Data Science) Department of CSE (Data Science)

ii
CERTIFICATE

iii
 DECLARATION

I hereby declare that results embodied in this 20CSD702-Project Work &

Internship “TRAINING PROGRAM RECOMMENDATION SYSTEM ” by me in
partial fulfillment of the award of Bachelor of Technology in Computer Science &
Engineering (Data Science) from Jawaharlal Nehru Technological University
Anantapur, Ananthapuramu and I have not submitted the same to any other
University/institute for award of any other degree.

Date :
Place : Madanapalle

PROJECT MEMBER
VUSA MAHENDRA - (21691A3252)

I certify that above statement made by the student is correct to the best of my
knowledge.

Date : Signature

iv
 TABLE OF CONTENTS

S.NO TOPIC PAGE NO.

1. INTRODUCTION 1
1.1 About Industry or Organization Details 2
1.2 My Personal Benefits 2
1.3 Objective of the Project 2
1.4 Limitations of Project 3

2. SYSTEM ANALYSIS 4
2.1 Introduction 5
2.2 Existing System 5
2.3 Disadvantages of Existing System 6
2.4 Proposed System 6
2.5 Advantages over Existing System 7

3. SYSTEM SPECIFICATION 9
3.1 Hardware Requirement Specification 10
3.2 Software Requirement Specification 10

4. SYSTEM DESIGN 12
4.1 System Architecture 13
4.2 Modules Flow Diagrams 14

5. IMPLEMENTATION AND RESULTS 17

5.1 Introduction 18
5.2 Implementation of Key Functions 18

v
 5.3 Method of Implementation (CODING) 19
5.4 Output Screens and Result Analysis 20

6. TESTING AND VALIDATION 22

6.1 Introduction 23
6.2 Design of Test Cases and Scenarios 23
6.3 Validation 24

7. CONCLUSION 27
7.1 Conclusion 28
REFERENCES 29

vi
 ABSTRACT

The "Training Program Recommendation System" is designed to support

faculty development at the National Institute of Technical Teachers Training and
Research (NITTTR). By providing personalized course recommendations, the
system assists faculty members in identifying relevant training programs based on
factors such as mode preference (online or offline), demographic attributes, and prior
course completion history. Leveraging collaborative filtering techniques, the system
generates peer-informed recommendations to promote shared learning and enhance
teaching and research capabilities.

The technical framework incorporates data from faculty and training program
datasets, with preprocessing steps to ensure data consistency. The recommendation
model combines memory-based and model-based collaborative filtering approaches,
utilizing K-Nearest Neighbors (KNN) and Singular Value Decomposition (SVD)
algorithms to deliver accurate and scalable recommendations. Performance is
assessed using precision, recall, MAE, and RMSE metrics. This system ultimately
aims to foster a continuous, targeted professional development culture within
NITTTR, supporting faculty expertise and enriching the learning environment.

vii
 List of Figures

S.NO Figure Name of the figure Page

No. Number
1 2.1.1 Recommendation structure 05
2 2.4.1 Proposed model 07
3 3.2.1 Required libraries 10
4 4.1.1 Architecture 13
5 4.2.1 Collaborative filtering 15
6 6.3.1 Performance metrics 24

viii
 LIST OF ABBREVIATIONS

NITTTR National institute of technical teachers training

and research
SVD Singular value decomposition
RAM Random access memory
KNN K-Nearest Neighbors algorithm
SSD Solid State Drive
MAE Mean Absolute Error
RMSE Root Mean Squared Error
SK-Learn Scikit-Learn

ix
 CHAPTER 1

INTRODUCTION

1
1. INTRODUCTION

1.1 About Industry or Organization Details:

The National Institute of Technical Teachers Training and Research

(NITTTR) is a premier organization focused on training technical educators and
developing educational frameworks for technical institutions nationwide. It supports
faculty development through specialized training programs that align with
technological advancements and teaching methodologies. NITTTR’s mission is to
continually enhance the quality of technical education, and this project aligns with its
goal of advancing faculty skill development through targeted training
recommendations.

1.2 My Personal Benefits:

During this internship, I gained hands-on experience in data processing,

machine learning model development, and collaborative filtering algorithms.
Working on this project improved my technical skills in programming, particularly
with Python and machine learning libraries like Scikit-Learn and Surprise.
Additionally, I learned to approach problem-solving analytically, gained insights into
data-driven decision-making, and understood the importance of personalized
recommendations in educational settings. These skills have broadened my
understanding of real-world applications of AI in academic and professional
development contexts.

1.3 Objective of the Project:

The objective of the Training Program Recommendation System project is

to develop a tailored recommendation solution to support the professional growth of
faculty members at the National Institute of Technical Teachers Training and
Research (NITTTR). This system aims to enhance the faculty's ability to select
relevant training programs by providing personalized course suggestions based on
each faculty member's preferences, including mode of training (online or offline),
demographic attributes, and prior course completion history. By fostering a more
guided selection process, the system seeks to align faculty training with individual
needs and professional development goals.

2
 Through collaborative filtering techniques, the system delivers peer-informed
recommendations that promote a collaborative learning environment within
departments. By analyzing training trends among colleagues in similar fields, it
encourages faculty members to pursue programs that align with both personal and
departmental objectives. Ultimately, the system aims to improve faculty expertise,
thereby enhancing their teaching and research capabilities while contributing to
NITTTR’s overarching goal of advancing educational excellence.

1.4Limitations of Project:

 The project is limited by several factors:

o Data Limitations: The dataset is dependent on available records of
faculty and program information, which may affect the diversity of
recommendations.
o Algorithm Constraints: Collaborative filtering algorithms, while
effective, may not fully account for personal preferences beyond those of
similar users.
o Scalability: Expanding the system to handle larger datasets may require
additional processing power and optimized algorithms.
o Cold Start Problem: For new users with limited data, recommendations
may initially be less accurate until sufficient usage data is accumulated.

3
 CHAPTER 2

SYSTEM ANALYSIS

4
2.1 Introduction

System analysis is a critical step in designing any effective solution, as it helps

in identifying and clarifying the system’s specific needs, requirements, and
limitations. For the Training Program Recommendation System at NITTTR,
system analysis involves a thorough review of both existing and potential solutions,
allowing for a clear understanding of current gaps in the faculty training selection
process. This analysis examines the current manual or semi-manual processes used by
faculty to choose training programs, which may rely on either personal selection or
administrative recommendations without leveraging data-driven personalization. The
aim is to define how an automated recommendation system can streamline these
processes, increase faculty engagement in professional development, and provide
value by tailoring recommendations to individual and departmental needs.

Figure 2.1.1 Recommendation structure

2.2 Existing System

NITTTR lacks a comprehensive automated system for recommending relevant

training programs to faculty members. The existing approach to course selection is
largely based on subjective preferences or guidance from administrative staff, without
significant use of data-driven methods. Faculty may choose courses based on general
information or popularity, which can result in training selections that are misaligned
with their actual learning requirements, skill gaps, or professional growth objectives.

5
Furthermore, there is little to no tracking of faculty development goals in the current
system, which means that choices may not consistently align with the institution’s
vision for targeted faculty growth and inter-departmental knowledge enhancement.

2.3 Disadvantages of Existing System

The lack of an automated recommendation system leads to several disadvantages:

 Limited Personalization: The current method fails to provide
recommendations tailored to the individual needs and professional
backgrounds of faculty members, resulting in a less effective learning
experience.
 Absence of Peer-Based Suggestions: Without a system that takes peer choices
into account, faculty miss out on training courses completed by their
colleagues, limiting the potential for collaborative learning and shared
development goals.
 Missed Opportunities for Professional Growth: The lack of a structured,
data-driven recommendation approach means that faculty may overlook
courses that could greatly benefit their careers. This leads to missed
opportunities to enhance skills and competencies that align with both their
individual goals and the organization’s objectives.
 Inconsistent Alignment with Departmental Goals: Since the existing system
does not prioritize departmental development targets, selected training
programs may not consistently support broader institutional goals or support
cross-functional learning within departments.

2.4 Proposed System

The proposed Training Program Recommendation System introduces a

data-driven, automated approach that utilizes collaborative filtering techniques to
offer personalized training program recommendations to faculty members. This
system is designed to analyze various factors, including each faculty member’s
course completion history, preferred learning mode, and demographic attributes, to
generate relevant suggestions that enhance individual and institutional growth. The
system applies both memory-based and model-based collaborative filtering methods
to suggest courses completed by faculty in similar roles, fostering a collaborative
learning environment. The objective is to streamline the course selection process by

6
aligning training recommendations with faculty members' professional goals, thereby
enriching the overall learning and teaching environment within NITTTR.

Figure 2.4.1 Proposed model

2.5 Advantages over Existing System

The proposed recommendation system offers several advantages over the current
system:
 Enhanced Personalization: By analyzing faculty data and individual
preferences, the system provides tailored recommendations, ensuring that each
faculty member receives training suggestions relevant to their unique learning

7
 and career development needs.
 Collaborative Learning Opportunities: The system promotes peer-based
recommendations, suggesting programs that other faculty members in the same
branch have completed. This encourages shared learning experiences, enabling
faculty to benefit from similar courses and to foster professional growth
through mutual knowledge sharing.
 Streamlined Faculty Development: With a data-driven recommendation
system, NITTTR can better align faculty training with departmental and
institutional objectives. By promoting training programs that support
departmental growth, the system ensures that professional development aligns
with institutional goals, promoting an integrated learning environment.
 Improved Course Discovery: The automated approach helps faculty discover
new and relevant courses that they might not have considered otherwise. By
recommending courses based on collaborative filtering, the system opens up
additional pathways for faculty to explore and expand their skill sets.
 Data-driven decision-making: The proposed system leverages insights from
faculty data to offer strategic recommendations, contributing to more informed
decision-making around professional development. This ensures that faculty
training choices are not only aligned with personal growth but also support
broader institutional strategies.

8
 CHAPTER 3

SYSTEM SPECIFICATION

9
3.1 Hardware Requirement Specification

 Processor: Intel Core i5 (8th generation or above) or AMD Ryzen 5.

 Memory (RAM): 8GB RAM.
 Storage: 256GB SSD.
 Graphics: Integrated graphics (sufficient for handling data processing and
model training tasks).

3.2 Software Requirement Specification

 Programming Language: Python 3.x.

 Development Environment:
o Jupyter Notebook: For interactive coding, data visualization, and model
testing.
o Git: For version control, to manage code versions and track project
changes.
 Libraries and Frameworks:

10
 Figure 3.2.1 Required libraries
o Data Processing Libraries:
 Pandas: For data manipulation and handling of structured
datasets.
 NumPy: For numerical operations and matrix manipulations
required in data preprocessing and similarity calculations.
o Machine Learning Libraries:
 Scikit-Learn: For data splitting, normalization, and machine
learning utilities.
 Surprise Library: For building collaborative filtering
recommendation algorithms like K-Nearest Neighbors (KNN) and
Singular Value Decomposition (SVD).
o Matplotlib / Seaborn: For data visualization, allowing visual insights
into data distributions and model performance.
 Database:
o Excel or SQLite or MySQL: For storing user and training program data.

11
 CHAPTER 4

SYSTEM DESIGN

12
4.1 System Architecture

The system architecture of the Training Program Recommendation System

includes several layers, each dedicated to a specific set of tasks, from data collection
and preprocessing to recommendation generation and evaluation. The architecture is
modular, allowing flexibility for further development and scalability.
Components of the System Architecture:
 Data Collection Layer:
o This layer collects data from two primary sources:
 Faculty Data: Includes demographic
information, completed courses, and user-
specific details.
 Training Program Data: Contains
information on available courses, including
program ID, department alignment, program
ratings, and instructor information.
 Data Preprocessing Layer:
o In this layer, raw data from the collection stage is
processed to make it suitable for analysis. Steps
include:
 One-Hot Encoding: Converts categorical
data (e.g., gender, mode preference) into a
numerical format.
 Normalization and Standardization:
Ensures numerical values are on a
comparable scale for effective model
performance.
 Data Splitting Layer:
o This stage divides the preprocessed data into
training and test sets to allow model evaluation and
validation. Common data splitting techniques used
are random splitting, chronological splitting, and stratified splitting.
Figure 4.1.1: Architecture

13
 Collaborative Filtering Models:
o This is the core layer where recommendation models are applied, using
both Memory-Based and Model-Based Collaborative Filtering
approaches:
 Memory-Based Filtering: Uses similarity calculations between
users or items to recommend relevant courses (User-based and
Item-based).
 Model-Based Filtering: Applies Singular Value Decomposition
(SVD) for dimensionality reduction, ensuring efficient
recommendations for larger datasets.
 Evaluation Layer:
o The output recommendations from the models are evaluated using
metrics like Precision, Recall, Mean Absolute Error (MAE), and Root
Mean Squared Error (RMSE). This ensures the system’s performance
meets expected standards.
 Output Layer:
o The final recommendations are provided to faculty members, offering
personalized course suggestions that align with their individual learning
preferences and professional development goals.

4.2 Modules Flow Diagrams

The module flow diagrams represent the sequential steps within each layer of
the system, providing a visual representation of the entire recommendation process.
Flow Diagram Descriptions:
 Data Collection Module:
o Inputs: Faculty data and training program data.
o Process: Collects, organizes, and stores user and program details.
o Output: Structured datasets ready for preprocessing.
 Data Preprocessing Module:
o Inputs: Raw data from the collection layer.
o Process: Applies one-hot encoding, normalization, and standardization
to prepare data for model training.
o Output: Clean, structured data ready for model input.

14
 Data Splitting Module:
o Inputs: Preprocessed data.
o Process: Divides data into training and testing sets for model validation,
ensuring data is representative of real-world scenarios.
o Output: Training and testing datasets for collaborative filtering models.
 Collaborative Filtering Models Module:

Figure 4.2.1 Collaborative Filtering

o Inputs: Training dataset.
o Process:
 User-Based Filtering: Finds similar users and recommends
courses based on shared interests.
 Item-Based Filtering: Identifies courses that are frequently
attended together and recommends accordingly.
 SVD Model: Reduces data dimensionality and generates accurate
recommendations for larger datasets.
o Output: Recommended courses for each faculty member.
 Evaluation Module:
o Inputs: Recommendations from collaborative filtering models.

15
 o Process: Assesses recommendation accuracy using metrics like
precision, recall, MAE, and RMSE.
o Output: Performance scores for each model.
 Recommendation Output Module:
o Inputs: Evaluated recommendations.
o Process: Presents final recommendations to faculty in a clear and user-
friendly manner.
o Output: Personalized course recommendations for faculty professional
development.

16
 CHAPTER 5

IMPLEMENTATION AND RESULTS

17
5.1 Introduction

The Implementation and Results phase is critical in transforming the project

design and theoretical framework into a functional system. This phase involves
developing and integrating key components such as data preprocessing, collaborative
filtering models, and evaluation metrics to achieve the project’s objective of
delivering accurate and personalized recommendations. By implementing
collaborative filtering techniques, the system can suggest relevant training programs
tailored to faculty members' profiles and professional needs. The results generated by
the system are then analyzed and validated to ensure accuracy and relevance.

5.2 Implementation of Key Functions

Key functions of the Training Program Recommendation System are categorized

as follows:
 Data Preprocessing:
o One-Hot Encoding and Normalization: These methods prepare
categorical features (e.g., gender, department) and standardize numerical
values (e.g., ratings) to improve model performance and compatibility.
o Data Splitting: The dataset is divided into training and testing sets to
validate the model’s accuracy and prevent overfitting.
 Collaborative Filtering Models:
o User-Based Collaborative Filtering: Calculates similarity scores
between users to recommend programs that similar users have
completed.
o Item-Based Collaborative Filtering: Identifies similarities between
items (programs) and recommends courses based on similar item ratings.
o Model-Based Collaborative Filtering (SVD): Singular Value
Decomposition is implemented for dimensionality reduction, allowing
the model to make accurate predictions with sparse data.
 Evaluation Metrics:
o Implemented metrics include Precision, Recall, Mean Absolute Error
(MAE), and Root Mean Squared Error (RMSE) to measure the
system’s recommendation accuracy.

18
5.3 Method of Implementation (Coding)

19
Below is a sample of the coding approach used to implement the core functions:

20
5.4 Output Screens and Result Analysis

The recommendation system provides a list of suggested training programs for

each faculty member based on the similarity scores computed. Below is a sample
output format:
Output:

Recommended courses for user 2412224195:

21
['GIS APPLICATIONS IN SMART CITY DEVELOPMENT',
'TRANSPORTATION PLANNING AND GREEN BUILDING', 'DISASTER
MANAGEMENT']
Recommended courses for user 2412224195:
[ 'GIS APPLICATIONS IN SMART CITY DEVELOPMENT',
'TRANSPORTATION PLANNING AND GREEN BUILDING', 'DISASTER
MANAGEMENT']

22
 CHAPTER 6

TESTING AND VALIDATION

6.1 Introduction

The Testing and Validation phase ensures that the system meets its functional

23
requirements, providing personalized training program recommendations that align
with user preferences and needs. This section covers the process of designing test
cases, evaluating model performance, and validating recommendations using
appropriate metrics. Effective testing and validation contribute to the reliability and
usability of the recommendation system in real-world scenarios.

6.2 Design of Test Cases and Scenarios

Test cases are designed to verify that the system performs accurately across
various functions and scenarios, including data processing, model performance, and
recommendation accuracy. Here are key test cases for the recommendation system:
1. Data Integrity Test:
o Objective: Ensure that data from both participants_df and dept_df is
correctly loaded, cleaned, and preprocessed.
o Test Steps: Verify that categorical data is encoded, numerical data is
normalized, and data types match across merged columns.
o Expected Outcome: Data integrity is maintained with no missing or
mismatched values.
2. Merge Validation Test:
o Objective: Test the successful merging of participants_df and dept_df
based on course_id, department, and courses_attended.
o Test Steps: Confirm that merged DataFrame includes only relevant
columns and matched rows from both datasets.
o Expected Outcome: Accurate merging without empty rows or incorrect
matches.
3. User-Based Collaborative Filtering Test:
o Objective: Verify that the user-based collaborative filtering model
recommends relevant courses based on user similarities.
o Test Steps: Select a test user, compare their actual course history to the
model’s recommendations, and verify that similar users’ choices
influence suggestions.
o Expected Outcome: Recommended courses align with the user’s
interests and prior course completion patterns.
4. Item-Based Collaborative Filtering Test:
o Objective: Ensure that item-based collaborative filtering accurately
recommends similar courses.

24
 o Test Steps: Check that the model identifies similar courses based on
other users’ ratings and course characteristics.
o Expected Outcome: Recommended courses have content or
characteristics similar to the user’s previously attended courses.
5. Model Evaluation Metrics Test:
o Objective: Confirm that model performance metrics (e.g., Precision,
Recall, MAE, RMSE) are correctly calculated.
o Test Steps: Run the model evaluation functions and check metric
outputs for each recommendation model.
o Expected Outcome: Evaluation metrics accurately reflect model
performance, with meaningful values for each metric.
6. Cold Start Test:
o Objective: Test system performance when recommending courses for
new users or users with limited data.
o Test Steps: Use a new faculty profile with minimal data and assess if the
system recommends general or popular courses from the same
department.
o Expected Outcome: The system provides relevant courses based on
general trends in the user’s department.

6.3 Validation

Figure 6.3.1 Performance metrics

The system’s recommendations are validated using several key performance
metrics:
1. Precision:
o Measures the proportion of relevant recommended items among all

25
 recommended items.
o Formula:

2. Recall:
o Measures the proportion of relevant recommended items among all
relevant items in the test set.

o Formula:

3. Mean Absolute Error (MAE):

o Measures the average absolute difference between predicted and actual
user ratings for recommended items.
o Formula:

26
4. Root Mean Squared Error (RMSE):
o Assesses the model’s accuracy by measuring the square root of the
average squared differences between predicted and actual ratings.
o Formula:

27
 5. Coverage:
o Measures the diversity of recommendations, indicating the proportion of
items in the dataset that are recommended.
o Formula:

These metrics provide a comprehensive understanding of the model’s ability to

deliver relevant, accurate, and varied recommendations. Regular validation helps
ensure the system maintains high performance as new data is introduced.

28
CHAPTER 7

CONCLUSION

29
7.1 Conclusion

The Training Program Recommendation System project successfully

developed a tailored solution to support faculty development at the National Institute
of Technical Teachers Training and Research (NITTTR). Using collaborative
filtering techniques, the system delivers personalized course recommendations based
on faculty preferences, past training history, and department alignment. This
approach promotes faculty engagement in relevant training programs, fostering
professional growth and supporting a collaborative learning environment. The
system’s design and implementation, from data preprocessing to model testing,
achieved high performance. Evaluation metrics such as precision, recall, MAE, and
RMSE demonstrated the system’s accuracy and effectiveness in generating
meaningful recommendations aligned with faculty needs.

This recommendation system significantly enhances NITTTR’s capability to

support its faculty by aligning training suggestions with individual and departmental
goals. By recommending peer-completed courses, the system encourages shared
learning and contributes to knowledge sharing across departments. Additionally, its
capability to provide recommendations for new or low-data users adds versatility.
Overall, the project aligns with NITTTR's mission to enhance teaching and research
quality through targeted professional development, creating a valuable tool that can
evolve with future data and user needs.

30
REFERENCES

Gediminas Adomavicius and Alexander Tuzhilin, "Recommender Systems:

Challenges and Research Opportunities," Communications of the ACM, 2005.

Yehuda Koren, Robert Bell, and Chris Volinsky, "Matrix Factorization Techniques
for Recommender Systems," IEEE Computer Society, 2009.

Alexandros Karatzoglou and Balázs Hidasi, "Deep Learning for Recommender

Systems," Proceedings of the RecSys Conference, 2015.

Shlomo Geva and Hananeh Ghasemzadeh, "A Survey of Collaborative Filtering

Techniques," ACM Transactions on Information Systems, 2016.

Claudio Carpineto and Giovanni Romano, "A Survey on User Profiling in

Recommender Systems," Springer Handbook of Research on User Modeling and
Profiling, 2018.

Paolo Cremonesi, Yehuda Koren, and Roberto Turrin, "Hybrid Recommender

Systems: Survey and Experiments," Springer Handbook on Recommender Systems,
2015.

F. Ricci, L. Rokach, and B. Shapira, "Introduction to Recommender Systems

Handbook," Springer Handbook on Recommender Systems, 2015.

31
Michael D. Ekstrand, John T. Riedl, and Joseph A. Konstan, "Collaborative Filtering
Recommender Systems," Foundations and Trends in Human-Computer Interaction,
2011.

32