Mini-Project – 2B Web based on ML

(ITM 601)
Cryptocurrency Price Predictor
T. E. Information Technology

By

Ashna Brito 02
Glenn Fernandes 17
Shubham Jadhav 22
Levin Nadar 37

Under the Guidance of

Prof. Sunantha Krishnan

Department of Information Technology

Don Bosco Institute of Technology
Academic year 2024-2025
 The Salesian Society’s
Don Bosco Institute of Technology
Vidyavihar Station Road, Mumbai - 400070

CERTIFICATE

This is to certify that the project entitled “Cryptocurrency Price Predictor” is a

bonafide work of “Ashna Brito, Glenn Fernandes, Shubham Jadhav, Levin Nadar ”
Roll Nos. 02,17,22,37” submitted to the University of Mumbai towards completion of
mini project work for the subject of Mini Project – 2B Web Based on ML (ITM 601).

Prof.Sunantha Krishnan
Project Supervisor

Prof. Sunantha Krishnan

HOD-IT

Examiners

1.

2.
 DECLARATION

We declare that this written submission represents our ideas in our own words and where others’
ideas or words have been included, we have adequately cited and referenced the original sources.
We also declare that we have adhered to all principles of academic honesty and integrity and have
not misrepresented or fabricated or falsified any idea/data/fact/source in our submission. We
understand that any violation of the above will be cause for disciplinary action by the Institute
and can also evoke penal action from the sources which have thus not been properly cited or from
whom proper permission has not been taken when needed.

02_Ashna Brito

17_Glenn Fernandes

37_Levin Nadar

22_Shubham Jadhav
 ABSTRACT

Cryptocurrencies are digital assets that have gained popularity due to their decentralized nature and
potential for high returns. However, their prices are highly volatile, which makes it difficult to predict their
future values using traditional methods. Accurate price prediction can help investors make better decisions
and reduce financial risks.

In this project, we aim to predict the closing price of a cryptocurrency using historical data and machine
learning techniques. We use features such as 'Open', 'High', 'Low', 'Volume', 'Year', 'Month', and 'Day' to
train our models. The data is preprocessed and scaled to improve the accuracy and efficiency of the models.

We experimented with several machine learning algorithms, including Neural Networks, Support Vector
Machines (SVM), Logistic Regression, and Random Forest. Each model was trained on past data and
evaluated using standard performance metrics such as MAE, MSE, RMSE, and MAPE. Among these, the
neural network showed the most promising results in learning complex patterns from the data.

This project shows how machine learning can be applied to financial data to make useful predictions. By
comparing different models, we provide insights into which algorithms work best for predicting
cryptocurrency prices. Our findings can help investors and analysts make data-driven decisions in the
fastchanging world of digital currencies.

The model is designed to predict the next day’s closing price using the most recent available data. By
inputting current market values, the system estimates the upcoming price based on learned patterns from
historical trends. This helps provide a short-term outlook that can assist in making informed trading
decisions without attempting long-term forecasting.

This project also emphasizes the value of comparing different machine learning models to identify the
best-performing approach. By testing Neural Networks, SVM, Logistic Regression, and Random Forest,
we could analyze their strengths and weaknesses in handling financial data. With continuous improvements
in data quality and model tuning, this system can serve as a helpful tool for understanding market behavior
in the fast-paced world of cryptocurrency.
 INDEX

Chapter Contents Page

No. No.

1 Introduction
1.1 Background (Description of the topic) 1
1.2 Need and Scope of the project 1
1.3 Objectives and Problem Statement 2
2 Literature Review 3
3 Proposed Work 4
3.1 Architectural Details 4
3.1.2 Data Collection 4

3.1.3 Data Preprocessing 5

3.1.4 Feature Selection 6

3.1.5 Train Test Split 7
3.1.6 Performance Evaluation 7
4 Implementation 8
4.1Algorithm Details 8
4.2 Dataset Details 10
4.3 Performance Metrics Details 11
5 Result and Discussions 17
6 Conclusion and Future Scope 18
References 18

Acknowledgement 19
 List of Abbreviations

Sr. No. Abbreviation Full Form

1 ML Machine Learning
2 SVM Support Vector Machine
3 NN Neural Network
4 ARIMA AutoRegressive Integrated Moving Average
5 GARCH Generalized Autoregressive Conditional
Heteroskedasticity
 List of Figures

Fig. No. Figure Name Page No.

3.1 Architecture Diagram 4

3.1.2 Data Collection 5
3.1.3 Data Preprocessing 5
3.1.4 Correlation Matrix Heatmap 6
3.1.5 Train Test Split 7
4.1.6 Testing 10
4.3.1 Neural Network Accuracy 13
4.3.2 Random Forest Accuracy 14
4.3.3 Support Vector Machine (SVM) Accuracy 15
4.3.4 Logistic Regression Accuracy 16
 Chapter 1 : Introduction
Our project focuses on predicting short-term cryptocurrency prices using historical market data and machine
learning. By analyzing key features such as price movements, trading volume, and time-based trends, the
model learns patterns that help estimate the next day’s closing price. This approach offers a data-driven
method to support informed trading decisions in a highly volatile market. The goal is to simplify complex
financial trends through predictive analytics, making it easier for users to understand market behavior and
reduce uncertainty.

1.1 Background
The cryptocurrency market is a rapidly growing sector that has garnered significant attention for its
potential to offer high returns and diversify investment portfolios. However, the market's volatility and
unpredictability make it challenging for investors to forecast price movements effectively. Traditional
methods of price prediction have often relied on technical analysis and expert opinions, which can be time-
consuming and subject to human error. With the rise of data-driven technologies, machine learning and
data analytics have emerged as powerful tools for improving price prediction accuracy. By leveraging
historical data and market trends, these technologies provide more efficient, objective, and precise insights,
allowing traders and investors to make informed decisions in an increasingly complex financial landscape.

1.2 Need and Scope of the project

The need for advanced technologies in financial markets, especially in cryptocurrency trading, has become
increasingly clear due to the market's volatility and complexity. Investors and traders face significant
challenges in making informed decisions due to unpredictable price movements and a lack of reliable
forecasting tools. Machine learning-powered price prediction systems can offer a more data-driven,
accurate approach to forecasting cryptocurrency prices. The scope of this project involves developing a
machine learning-based model that analyzes historical market data, such as price movements, trading
volume, and time-related patterns, to predict the next day's closing price for cryptocurrencies. This system
aims to assist traders in making better-informed, data-driven decisions, ultimately contributing to more
efficient and effective trading strategies in the dynamic cryptocurrency market.

1.3 Objectives and Problem Statement

The primary objective of this project is to develop an efficient and accurate cryptocurrency price prediction
system using machine learning techniques. The system aims to provide traders and investors with
datadriven insights that can help them make informed decisions about market trends, improving their
ability to predict short-term price movements. The problem statement centers around the complexities of
cryptocurrency price prediction, considering factors such as market volatility, trading volume, and external
influences that affect price behavior. Inaccurate price predictions can lead to poor investment decisions,
financial losses, and missed opportunities. By addressing these challenges through advanced machine
learning models, this project aims to provide a reliable solution for predicting cryptocurrency prices,
ultimately contributing to more informed trading strategies in a volatile market.

1
 Chapter 2 : Literature Review

Recent years have witnessed a growing interest in applying machine learning techniques to predict
cryptocurrency prices due to the highly volatile and non-linear nature of the market. Several studies have
explored different approaches to tackle the challenges associated with crypto price forecasting.

A 2021 study titled Cryptocurrency Price Prediction Using Traditional Statistical and Machine Learning
Techniques compared classical models like ARIMA and GARCH with machine learning algorithms such
as SVM, ANN, and LSTM. The findings indicated that deep learning models were more effective in
identifying complex, non-linear patterns in price movements. However, the study also highlighted
limitations in adapting to unpredictable events such as regulatory changes or global economic shocks,
which are not easily captured through historical data alone.

In 2022, the paper Cross Cryptocurrency Relationship Mining for Bitcoin Price Prediction introduced the
C2RM module, which utilized dynamic time warping (DTW) to discover synchronous and asynchronous
relationships between Bitcoin and Altcoin prices. This model improved prediction accuracy by capturing
price correlations across multiple cryptocurrencies. Despite its strength in analyzing inter-coin trends, the
model lacked consideration of external influences such as geopolitical events or social media sentiment.

Another 2022 study, Time Series Analysis of Blockchain-Based Cryptocurrency Price Changes, applied
LSTM networks to historical data for future price prediction. While effective in modeling temporal
dependencies and reducing prediction errors, the model struggled with capturing sudden market
movements like crashes or surges, due to its reliance on past patterns and lack of real-time data integration.

The 2023 research Modelling Determinants of Cryptocurrency Prices: A Bayesian Network Approach
offered a more comprehensive approach by incorporating social media sentiment (e.g., Twitter activity)
and macroeconomic indicators into a causal model. While this improved the contextual understanding of
price influences, the model faced challenges in filtering out misinformation and dealing with the noise
inherent in social media data.

2
 Chapter 3 : Proposed Work

Our proposed work focuses on collecting extensive cryptocurrency market data and applying machine
learning techniques to build predictive models for price analysis. These models will be trained and
validated using historical price trends, trading volumes, and other relevant indicators to ensure accuracy
and robustness.

3.1 Architectural Details

Fig. 3.1 – Architecture Diagram

3.1.2 Data Collection

Data collection is a foundational step in any data science project. For our cryptocurrency price
prediction system, we compiled a comprehensive dataset containing historical price data of
various cryptocurrencies such as Bitcoin, Ethereum, and others. The dataset includes key features
like open, close, high, and low prices, trading volume, market capitalization, and timestamps. In
some cases, external factors such as global economic indicators and social media sentiment were
also incorporated to enrich the dataset.
The data was sourced from Yahoo finance ensuring its reliability and relevance. Each entry
represents a snapshot of market behavior at a specific point in time, making the dataset a crucial
asset for training and evaluating machine learning models that aim to predict future price trends.

3
 Fig. 3.1.2 – Data Collection
3.1.3 Data Preprocessing
Data preprocessing plays a vital role in preparing raw financial data for effective analysis and
modeling. For this project, we performed the following preprocessing steps:
Handling Missing Values: We checked for any missing entries in key columns such as Open, High,
Low, Close, Volume, and date components using df.isnull().sum(). The dataset was complete, with no
null values detected, ensuring robust data for training and testing.
Duplicate Detection: To ensure data quality, we identified and removed any duplicate records using
df.duplicated().sum(). The absence of duplicates helped prevent data redundancy and maintained
dataset integrity.
Feature Engineering: From the Year, Month, and Day columns, we generated a unified Date feature to
maintain proper chronological order, crucial for time-series analysis. This also enabled easy
visualization of temporal trends.

Fig. 3.1.3 – Data Preprocessing

4
3.1.4 Feature Selection

Feature Selection for Cryptocurrency Price Prediction:

Correlation Analysis: We performed a correlation analysis using heatmaps to identify
relationships between the features (e.g., 'Open', 'High', 'Low', 'Close', 'Volume') and
the target variable (price or price change). Highly correlated features were prioritized
as they provide valuable information for the model and can improve its predictive
power.
Domain Knowledge: Domain knowledge of cryptocurrency markets was applied to
select features that influence cryptocurrency price movements. This includes technical
indicators such as trading volume, previous price trends ('Open', 'Close', 'High', 'Low'),
and market sentiment, which can be derived from features like social media data or
cryptocurrency news sentiment. Historical price data and trading volume are critical
for building a reliable prediction model.

Fig. 3.1.4 – Correlation Matrix Heatmap

5
 3.1.5 Train Test Split

The train-test split is essential for model evaluation and validation. We divided the dataset into
training and testing sets using train_test_split from sklearn.model_selection:
Training Set (80%): The training set comprised 80% of the data and was used to train machine
learning models on historical data.
Testing Set (20%): The remaining 20% of the data constituted the testing set, which served as
unseen data for evaluating model performance. Testing on unseen data helps assess how well the
model generalizes to new instances and prevents overfitting.

Fig. 3.1.5 – Data Split

3.1.6 Performance Evaluation

Model Training: We trained multiple machine learning algorithms, including Neural Networks (NN),
Support Vector Machines (SVM), Logistic Regression, and Random Forest using historical cryptocurrency
data. Features such as 'Open', 'High', 'Low', 'Close', and 'Volume' were used to predict the future
cryptocurrency prices.

Prediction and Evaluation: After training the models, we made predictions on the test dataset and
compared these predicted values with the actual market values. Evaluation metrics such as Mean Squared
Error (MSE), Mean Absolute Error (MAE), R-squared, and Root Mean Squared Error (RMSE) were
computed using functions from libraries like sklearn.metrics. These metrics helped assess the accuracy
of the models in predicting future prices.

Comparison and Selection: We compared the performance of each model based on the evaluation metrics.
The models that showed lower RMSE and MAE, as well as higher R-squared values, were considered better
for accurate price prediction. The most effective model(s), such as Random Forest or NN, were selected
based on their predictive accuracy, reliability, and ability to handle the market data's complexities.

6
 Chapter 4 : Implementation

4.1 Algorithm Details

1. Data Collection and Preprocessing:

Gather Data: Collect historical cryptocurrency data that includes features such as 'Open', 'High', 'Low',
'Close', 'Volume', 'Year', 'Month', and 'Day'.
Preprocessing:
Handling Missing Values: Check for missing values and handle them by either imputation or
removal, ensuring completeness.
Feature Scaling: Scale numerical features (like 'Open', 'High', 'Low', 'Volume') using
techniques such as Min-Max scaling or Standardization to improve model convergence.
Date-Time Processing: Convert the 'Year', 'Month', and 'Day' columns into a datetime format
and extract additional temporal features like day of the week, or whether it’s a holiday.
Encoding Categorical Variables: If there are any categorical variables, encode them
appropriately using techniques like One-Hot Encoding or Label Encoding.

2. Feature Selection:
Identify Key Features: Based on the problem at hand, focus on features that most likely affect the
cryptocurrency price prediction, such as:
'Open', 'High', 'Low', 'Close', 'Volume': These are crucial in predicting the future price
trends.
Time-based Features: Extract temporal features from 'Year', 'Month', and 'Day' to account
for seasonal or cyclical behavior.

Neural Network (NN) Algorithm Training:

Model Structure:
Use the Neural Network (NN) algorithm, a powerful tool for capturing non-linear relationships
in the data.
Train the NN model using selected features such as 'Open', 'High', 'Low', 'Close',
'Volume', and any additional technical indicators (e.g., moving averages, RSI, etc.).
The NN will consist of multiple layers, including an input layer, hidden layers, and an output
layer, with activation functions
Training Process:
The model learns by adjusting weights and biases to minimize the error (using
backpropagation).
Target Variable: The target could be the future cryptocurrency price or the price change
(up/down).

7
Random Forest Algorithm Training:
Ensemble Learning:
Implement the Random Forest algorithm by constructing an ensemble of decision trees. Each
tree is trained on a subset of the data to make individual predictions.
The Random Forest aggregates the predictions from multiple trees to enhance predictive
accuracy and reduce overfitting.
Training Process:
Input Features: The model is trained using features such as 'Open', 'High', 'Low', 'Close',
'Volume', and technical indicators.
The Random Forest model creates multiple decision trees, each trained on a random subset of
the data, and then aggregates the results to improve the final prediction's robustness and
stability.

Logistic Regression Algorithm Training:

Binary Classification:
Use Logistic Regression for binary classification tasks, such as predicting whether the price
will go up (1) or down (0).
The algorithm learns a linear relationship between input features and the probability of the
binary outcome.
Training Process:
Input Features: Features like 'Open', 'High', 'Low', 'Close', 'Volume', and indicators such
as RSI or moving averages.
Output: The model predicts the probability of the price increasing or decreasing. o The
model uses a logistic function (sigmoid) to output values between 0 and 1, which can be
interpreted as the likelihood of price movement.

Support Vector Machine (SVM) Algorithm Training:

Hyperplane Separation:
o Use SVM for classification tasks, such as predicting price direction (up/down). SVM
aims to find the optimal hyperplane that maximizes the margin between the different classes
(price rise vs. fall).
Training Process:
Input Features: Features like 'Open', 'High', 'Low', 'Close', 'Volume', and technical
indicators.
Output: The model classifies the data into two classes (e.g., rise or fall in price). o The
SVM finds the optimal hyperplane by maximizing the margin between the two classes, and
in non-linear scenarios, it uses kernels (e.g., RBF kernel) to transform data into higher
dimensions.

8
 1. Model Evaluation:
• Evaluate the performance of each trained model using metrics such as accuracy, precision,
recall, and F1-score.

2. Integration and Deployment:

Integrate the trained NN, Random Forest, Logistic Regression, and SVM models.

Fig. 4.1.6 - Testing

4.2 Dataset Details

Dataset Features
The dataset includes several key features that are crucial for cryptocurrency price prediction:
Date: The specific date for each record. Used for time series analysis and model indexing.
Open: The price of the cryptocurrency at the beginning of the trading day.
High: The highest recorded price of the cryptocurrency during the trading day.
Low: The lowest recorded price of the cryptocurrency during the trading day.
Close: The price of the cryptocurrency at the end of the trading day. Often used as the primary target
variable for prediction.
Adj Close (Adjusted Close): Adjusted closing price that accounts for splits, dividends, and other
corporate actions. It provides a more accurate reflection of the cryptocurrency’s value over time.
Volume: The number of units of cryptocurrency traded during the day, indicating market activity and
liquidity.
These features serve as the basis for various financial analyses including trend detection, volatility
measurement, and price forecasting.

9
Target Variable
Depending on the project objective, the target variable can be framed in two ways:
Regression Task: Predicting the next day's closing price based on past trends and indicators.
Classification Task: Predicting the direction of price movement (i.e., whether the price will go
up or down the next day).
The flexibility of the dataset makes it suitable for both supervised learning tasks.

Dataset Size and Integrity

The dataset contains 2,412 entries, with each row representing one day of trading data.
There are no missing values or duplicate entries in the dataset.
Data is chronologically ordered and suitable for time series forecasting and sequence modeling. This
ensures a high level of integrity, which is crucial for accurate model development and evaluation.

Data Types
Numerical Features: Open, High, Low, Close,Volume.
Temporal Feature: Date.
Derived/Categorical Feature (optional): Price Direction (Up / Down) – can be created using the
difference between today’s and tomorrow’s close price.
These data types allow for both time series analysis and standard supervised learning models.

10
4.3 Performance Metrics Details

Algorithm: Neural Network Accuracy:

The NN model achieved an accuracy of approximately 10.65%, indicating that its

predictions are correct only about 10.65% of the time. This low accuracy signifies
that the model struggles to predict values.The Mean Absolute Error (MAE) of
1202.03, Mean Squared Error (MSE) of 2,186,021.58, and Root Mean Squared
Error (RMSE) of 1478.52 reflect significant discrepancies between predicted and
actual values, pointing to areas in need of improvement.

Fig. 4.3.1 – Neural Network Accuracy

11
Algorithm : Random Forest
The Random Forest model excelled in the cryptocurrency price prediction task, achieving a remarkable
accuracy rate of 90%. This high level of accuracy highlights the model’s effectiveness in handling complex
financial classification problems and predicting market trends based on historical price patterns and trading
volume. With strong performance across evaluation metrics such as precision, recall, and F1-score, SVM
demonstrates robustness and consistency in identifying subtle market signals. The cross-validation results
further reinforce the model’s reliability, confirming its potential as a powerful tool for informed decision-
making

12
Algorithm : Support Vector Machine (SVM) Accuracy
The Support Vector Machine (SVM) algorithm achieved a commendable accuracy rate of 97.7%
in crop categorization, highlighting its effectiveness in handling classification tasks based on
intricate agricultural features. With high precision and recall values across various crop categories,
SVM demonstrates its ability to accurately classify crops even in complex scenarios. The F1-score
metrics indicate a balanced performance in terms of precision and recall, ensuring reliable
predictions. The consistent cross-validation scores further validate the algorithm's robustness,
making it a valuable tool for crop recommendation.

13
Logistic Regression Accuracy:
The Logistic Regression algorithm achieved an accuracy rate of 48.2% in predicting cryptocurrency trends
based on various factors. This accuracy suggests that the model is not significantly outperforming random
chance, highlighting the challenges of predicting volatile markets like cryptocurrencies. The precision,
recall, and F1-score metrics were somewhat below expectations, indicating that the model struggled to
make consistent, accurate predictions in this complex domain. This suggests that the algorithm may benefit
from more sophisticated features, better data preprocessing.

Fig. 4.3.4 – Logistic Regression’s Accuracy

14
Support Vector Machine (SVM) emerged as the best-performing model in the cryptocurrency price
prediction task. Its ability to find the optimal hyperplane in high-dimensional space allows it to model
complex nonlinear patterns in financial data. SVM with a suitable kernel (like RBF) efficiently captured
the volatile trends in the crypto market while minimizing overfitting. Despite requiring proper tuning and
being computationally intensive, its superior generalization capability gave it the edge over other models.

Random Forest followed closely behind SVM in performance. As an ensemble learning method, it
leveraged the power of multiple decision trees to reduce variance and improve accuracy. Its strength lies
in handling non-linear relationships and missing data well, which are common in crypto datasets. However,
it fell slightly behind SVM due to its tendency to become less precise with very volatile price swings, as it
may overfit to noise if not properly controlled.

Neural Networks (NNs), while powerful, ranked third in this comparison. They are capable of learning
complex and abstract data representations, making them suitable for time-series prediction. However, NNs
require a large amount of data and careful tuning of architecture and hyperparameters. In this project,
limited data and high price volatility posed challenges in achieving consistent and stable results, leading
to slightly lower prediction accuracy compared to SVM and Random Forest.

Logistic Regression performed the worst in the context of price prediction. Unlike the other models,
Logistic Regression is fundamentally a classification algorithm, not regression. It predicts binary or
categorical outcomes, making it ill-suited for forecasting continuous numerical values like cryptocurrency
prices. While it can be adapted into regression (i.e., Linear Regression), in its original form, it lacks the
flexibility and complexity needed for financial forecasting tasks.

In conclusion, SVM was the most effective model for predicting cryptocurrency prices, with Random
Forest and Neural Networks as solid alternatives depending on data availability and project goals.
Logistic Regression, on the other hand, should be avoided for continuous price prediction tasks due to its
classification-oriented nature. For future improvements, combining SVM or Random Forest with LSTM-
based models or technical indicators could yield even more robust and accurate results.

15
 Chapter 5 : Results and Discussions

The Cryptocurrency Price Predictor achieved commendable results across various machine learning
algorithms, confirming the richness of the dataset and the models’ ability to accurately forecast crypto
market behavior. The models were assessed based on their effectiveness in classifying whether the price
of a cryptocurrency would increase or decrease the following day—an insight of high value to traders and
analysts.

Among the algorithms evaluated, Support Vector Machine (SVM) delivered the best performance across
all key metrics, including accuracy, precision, recall, and F1-score. Its ability to handle high-dimensional
data and clearly define decision boundaries made it particularly effective in capturing the subtle and
complex relationships inherent in cryptocurrency market trends. SVM’s robustness against overfitting and
its capacity to maintain strong generalization across different marketegments and volatility patterns made
it the most dependable model for prediction tasks in this domain.

Random Forest followed closely, offering competitive accuracy and high reliability. The model’s ensemble
structure, built on multiple decision trees, allowed it to identify non-linear patterns while remaining
resilient to overfitting. It performed consistently well across both training and testing datasets and had the
added advantage of interpretability through feature importance analysis. This transparency made it easier
to understand which input variables, such as trading volume, price range, and previous day trends, were
most influential in determining prediction outcomes.

Neural Networks also showed considerable promise, especially in modeling complex and non-linear data.
Their ability to learn intricate dependencies was particularly beneficial in understanding volatile financial
time series. However, they required greater computational resources and more fine-tuned training
procedures. Their “black-box” nature, combined with longer training times and limited interpretability,
presented some challenges when compared to tree-based models.

Logistic Regression, although straightforward and computationally efficient, fell short when it came to
capturing the deeper, non-linear patterns present in the cryptocurrency market data. It served well as a
baseline model but lacked the predictive depth and adaptability exhibited by the other algorithms tested.

After comprehensive comparison and analysis, Support Vector Machine was selected as the final model
due to its superior predictive performance, resilience to market fluctuations, and consistent results across
different evaluation criteria. While Random Forest and Neural Networks hold significant potential for
realworld deployment, especially when interpretability or scalability are priorities, the SVM model
demonstrated the most accurate and stable performance for the current scope of this project, making it the
most suitable algorithm for cryptocurrency price prediction applications.

16
 Chapter 6 : Conclusion and Future Scope

3.1 Conclusion

In conclusion, the implementation of machine learning algorithms, particularly Support Vector Machines
(SVM), for cryptocurrency price prediction has yielded highly accurate results in forecasting price
movements. This research confirmed the ability of SVM to effectively classify whether the price of a
cryptocurrency would rise or fall the following day, offering valuable insights for traders and investors
navigating the volatile digital asset market. SVM consistently delivered the highest accuracy across various
evaluation metrics, closely followed by Random Forest, Neural Networks, and Logistic Regression.
The model’s success lies in its ability to handle high-dimensional financial data and detect subtle patterns
in noisy market environments. Random Forest also proved effective due to its ensemble nature and
interpretability, while Neural Networks demonstrated potential for capturing complex nonlinear
relationships, albeit at the cost of interpretability and longer training times. Logistic Regression, though
straightforward and computationally efficient, lacked the capacity to model the intricacies of crypto market
dynamics.
Overall, the SVM model emerged as the most suitable for deployment due to its superior accuracy,
generalization ability, and resilience to market fluctuations. Its implementation can aid in developing
reliable trading signals, enhancing portfolio strategies, and informing decision-making in real-time
financial applications.

Future work can further elevate the effectiveness of this system. First, incorporating sentiment analysis
from news headlines and social media could help capture market sentiment and investor behavior.

Additionally, integrating real-time data streaming and online learning models could enable dynamic
updates to predictions as new data arrives. Expanding the dataset to include macroeconomic indicators,
regulatory changes, and global market trends would improve model generalization across different
economic conditions. Finally, deploying the model through interactive dashboards or mobile applications
can make these insights more accessible to retail traders and financial analysts alike.
By pursuing these enhancements, machine learning-based cryptocurrency prediction tools can become
powerful assets in modern finance—bridging the gap between data science and actionable investment
strategies.

17
 References

[1] A. M. Khedr, I. Arif, P. V. Pravija Raj, M. El‐Bannany, S. M. Alhashmi, and M.

Sreedharan, (2021) Cryptocurrency price prediction using traditional statistical and
machine-learning techniques: A survey
[2] Li, P., Gong, S., Xu, S., Zhou, J., Shanqing, Y., & Xuan, Q. (2022). Cross Cryptocurrency
Relationship Mining for Bitcoin Price Prediction.
[3] J. Fleischer, G. von Laszewski, C. Theran, and Y. J. P. Bautista, (2022) Time series
analysis of blockchain-based cryptocurrency price changes,
[4] Amirzadeh, R., Nazari, A., Thiruvady, D., & Ee, M. S. (2023). Modelling Determinants
of Cryptocurrency Prices: A Bayesian Network Approach.
[5] Wu, J., Zhang, X., Huang, F., Zhou, H., & Chandra, R. (2024). Review of Deep Learning
Models for Crypto Price Prediction: Implementation and Evaluation.R. Ghadge, J.
Kulkarni, P. More, S. Nene, and R. Priya, “Prediction of crop yield using machine
learning,” Int. Res. J. Eng. Technol.(IRJET), vol. 5, 2018.

18
 Acknowledgment
We are thankful to our college Don Bosco Institute of Technology for giving us this chance to
gain exposure in solving real world problems and acquire practical knowledge and skill sets that
will prove to be very crucial to our long-term career prospects.

We would take this opportunity to express our sincerest gratitude to our panel members Prof.
Sunantha Krishnan for her encouragement and guidance that she gave during our term
presentations . This project, and the research that we undertook, could not have been realized
without the utmost support of our Project Guide Prof. Sunantha Krishnan, who guided us every
step of the way, starting from the conception of the project, right up to the execution of the
finished solution

19