Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

PERFORMANCE ANALYSIS OF

MACHINE LEARNING
ALGORITHMS FOR
HANDWRITTEN DIGIT
RECOGNITION
AGENDA

 Introduction
 Objective
 Literature survey
 Project flow diagram
 Methodology
 Results
 Results and analysis
Introduction

 Handwritten Digit Recognition is the capacity of a computer to interpret the manually


written digits from various sources
 Machine Learning provides various methods through which human efforts can be reduced
in recognizing the manually written digits.
 The main objective was to actualize a pattern characterization method to perceive
the hand-written digits provided in the MINIST data set of images of handwritten digits (0‐9).
Objective

 To create a representation of isolated handwritten numbers that can be recognized


effectively.
 In any such recognition systems, addressing feature extraction and choosing the
classification approach would be the most important problem.
 The proposed project will try to address both accuracy and time complexity. To attain
the maximum accuracy in the recognition process by using the Machine Learning (ML)
model.
 This work is being done as a first attempt, and the main goal of this article is to make
handwritten numeral identification easier.
Literature Surveys
SI.N TITLE PUBLISHE AUTHORS METHODOLOGY ACCURACY RESULT
O D YEAR
1 Recognition of 2019 Fathma • CNN to classify The maximum Hence, this
Handwritten Digit Siddique; handwritten digits training accuracy proposed
using Convolutional using various numbers 100% and method of
Neural Network in Shadman of hidden layers and maximum CNN is more
Python with Tensor Sakib; epochs and to make validation efficient than
flow and the comparison accuracy 99.92% the other
Comparison of Md. Abu between the both at epoch existing
Performance for Bakr accuracies 15. The overall method for
Various Hidden Siddique • They performed their performance of digit
Layers experiment using the network is recognition.
Modified National found 99.21%.
Institute of Standards Moreover, the
and Technology overall loss
(MNIST) dataset. ranged from
Further, the network 0.026303 to
is trained using 0.049449
stochastic gradient
descent and the back-
propagation
algorithm.
2 Performance Evaluation 2019 Soufiane They exploited four KNN-99% Deep neural
of Machine Learning Hamida; machine learning Decision networks is the
Algorithms in Bouchaib algorithms (KNN, DNN, Tree-98% most accurate
Handwritten Digits Cherradi; Decision Tree and SVM) SVM-98% compared to the
Recognition Abdelhadi to predict handwritten DNN-100% other classifiers
RAIHANI; digits. The training and studied in this
Hassan testing data were paper.
OUAJJI extracted from the
MNIST digit database
containing pre-
processed images.
3 An efficient and 2019 Saqib Ali, The present research accuracy up The experimented
improved scheme for Zeeshan employed convolutional to 99.21% results demonstrate
handwritten digit Shaukat, neural network as that the proposed
recognition based on Muhammad classifier, MNIST as CNN framework for
convolutional neural Azeem, dataset with suitable MNIST dataset
network Zareen parameters for training exhibits high
Sakhawat, and testing and DL4J performance in
Tariq framework for terms of time and
Mahmood & handwritten digit accuracy as
Khalil ur recognition compared to
Rehman previously proposed
systems.
4 Handwritten digits 2019 Tsehay In this paper we have used The machine was The experiment
recognition with Admassu the standard Kaggle digits trained with a result shows that
decision tree Assegie, dataset for recognition of Kaggle dataset the decision tree
classification: a Pramod handwritten digits using a which contains classifier is
machine learning Sekharan decision tree classification 42000 rows and effective in
approach Nair approach. And we have 720columns and recognition of
evaluated the accuracy of result shows 83.4% handwritten
the model against each accuracy. digits.
digit from 0 to 9

5 Handwritten Digit 2020 Rohan In this paper, we have The accuracy is The output
Recognition Using Sethi; applied the KNN further obtained is in the
Machine Learning Ila Classification Algorithm to calculated, for form of 28x28
Kaushik support the the process. And, matrix, after the
implementation of our hence the images completion of
project work, “Hand- are classified training the model
written digit based on KNN in supervised
classification”. algorithm where machine learning
in this paper we algorithm and the
proposed the system predicts
value of k=5, the values of
according to their digits correctly.
project.
6 Handwritten 202 Rabia In this study, handwriting digit SVM-90% Hence, the K-
Digit 1 KARAKAYA, recognition process has been Decision Tree-87% means algo and
Recognition Serap done with algorithms having Random Forest- KNN gives better
Using Machine KAZAN different working methods. 97% accuracy when
Learning These algorithms are Support ANN-97% compared to
Vector Machine (SVM), Decision KNN-98% other algorithms
Tree, Random Forest, Artificial K-Means
Neural Networks (ANN), K- Algorithm-98%
Nearest Neighbor (KNN) and K-
Means Algorithm.
Tests were performed on the
MNIST handwritten digit data set
Data flow diagram
Results

1. For CNN, the accuracy obtained was 99%


2. For SVM, the accuracy obtained was 91%
Result analysis

 From the results it can be concluded that CNN is best for image classification than
SVM. In CNN has greater accuracy of 99% using 10 epochs in comparison with other
implemented models on testing samples (10,000 images from MNIST Dataset) and the
CNN model is trained on training samples (60,000 images from MNIST Dataset).
 In handwritten digit or numeral recognition, this type of improved precision will work
together to help the machine execute more efficiently by speeding up the performance
of the machine. And training time is also great for CNN (i.e.,10 minutes).
 Convolutional Neural Networks (CNN) may be able to improve image resolution and
noise processing by providing improved performance. So, Graphical User Interface is
implemented using CNN rather than SVM. .As SVM uses a technique called the kernel
trick to transform your data and then based on these transformations it finds an optimal
boundary between the possible outputs.
THANK YOU

You might also like