Garuda - Garba Rujukan Digital

Time Series Analysis of Bitcoin Prices Using ARIMA and LSTM for Trend Prediction Berlilana; Wahid, Arif Mu’amar
Journal of Digital Market and Digital Currency Vol. 1 No. 1 (2024): Regular Issue June 2024
Publisher : Bright Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47738/jdmdc.v1i1.1

This study investigates the efficacy of ARIMA and LSTM models in predicting Bitcoin prices, emphasizing the importance of accurate price prediction for trading, risk management, and investment strategies in the volatile cryptocurrency market. The objectives are to analyze Bitcoin prices to identify underlying patterns and trends, compare the predictive performance of ARIMA and LSTM models, and provide insights into their practical applications for Bitcoin price prediction. A comprehensive dataset of Bitcoin prices from January 1, 2011, to December 31, 2023, sourced from CoinMarketCap, was used. Data preprocessing included handling missing values, removing duplicates, achieving stationarity through differencing, and normalizing data using MinMaxScaler. The ARIMA model's best-fitting parameters were identified using ACF and PACF plots, and it was trained with the statsmodels library. The LSTM model involved data preparation through windowing and train-test splitting, constructing a neural network with LSTM layers, and training using TensorFlow/Keras. Evaluation metrics included Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE), with comparisons based on accuracy and computational efficiency. The ARIMA model demonstrated impressive performance with an MAE of 2.308392356829177e-215 and an RMSE of 0.0, indicating a near-perfect fit to the training data. The LSTM model achieved an MAE of 0.00021804577826689423 and an RMSE of 0.00021916977109865863, showing robust performance in handling nonlinear and long-term dependencies. The ARIMA model excelled in computational efficiency with a training time of 2.548070192337036 seconds and a prediction time of 0.0009970664978027344 seconds, while the LSTM model required 378.69622468948364 seconds for training and 0.6859967708587646 seconds for prediction. The results highlight ARIMA's effectiveness in capturing linear trends and its suitability for short-term trading strategies, while LSTM is better for long-term investment strategies due to its ability to model complex patterns. Despite potential overfitting in ARIMA and high computational demands for LSTM, the study suggests exploring hybrid models, incorporating additional data sources, and developing advanced techniques to enhance predictive accuracy in future research.

Analysis of Apriori and FP-Growth Algorithms for Market Basket Insights: A Case Study of The Bread Basket Bakery Sales Hery; Widjaja, Andree E.
Journal of Digital Market and Digital Currency Vol. 1 No. 1 (2024): Regular Issue June 2024
Publisher : Bright Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47738/jdmdc.v1i1.2

Market basket analysis is a crucial technique in retail for uncovering associations between items frequently purchased together. This study aims to compare the effectiveness of the Apriori and FP-Growth algorithms using sales data from "The Bread Basket" bakery, comprising 20,507 transactions. Key variables include TransactionNo, Items, DateTime, Daypart, and DayType. The data underwent preprocessing steps, including cleaning, tokenization, and feature extraction using TF-IDF. The Apriori and FP-Growth algorithms were implemented with hyperparameter tuning and an 80/20 training/testing split. Performance metrics were evaluated, revealing that Apriori had an execution time of 4.08 seconds and memory usage of 45.36 MiB, whereas FP-Growth exhibited an execution time of 4.15 seconds and significantly lower memory usage at 0.08 MiB. The quality of the association rules was assessed by metrics such as support, confidence, and lift. For example, the Apriori algorithm generated the rule {Alfajores} -> {Coffee} with support 0.018885, confidence 0.520000, and lift 1.087090, while FP-Growth produced the rule {Scone} -> {Coffee} with support 0.017829, confidence 0.519231, and lift 1.085482. FP-Growth generally outperformed Apriori, particularly in memory efficiency, due to its use of the FP-tree data structure, which reduces the need for multiple database scans. The practical implications for "The Bread Basket" bakery include optimizing product placement and inventory management based on the identified associations, such as placing Coffee near Cake or Medialuna to encourage complementary purchases. The study concludes that while both algorithms effectively generate meaningful association rules, FP-Growth's superior memory efficiency makes it more suitable for large datasets. Limitations include data quality and the study's scope, confined to a single bakery. Future research should explore hybrid approaches, real-time data analysis, and applications across different retail sectors to enhance market basket analysis techniques further.

Comparison of K-Means and DBSCAN Algorithms for Customer Segmentation in E-commerce Paramita, Adi Suryaputra; Hariguna, Taqwa
Journal of Digital Market and Digital Currency Vol. 1 No. 1 (2024): Regular Issue June 2024
Publisher : Bright Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47738/jdmdc.v1i1.3

Customer segmentation is crucial for e-commerce businesses to effectively target and engage specific customer groups. This study compares the effectiveness of two popular clustering algorithms, K-Means and DBSCAN, in segmenting e-commerce customers. The primary objective is to evaluate and contrast these algorithms to determine which provides more meaningful and actionable customer segments. The methodology involves analyzing a comprehensive e-commerce customer dataset, which includes various features such as customer ID, gender, age, city, membership type, total spend, items purchased, average rating, discount applied, days since last purchase, and satisfaction level. Initial data preprocessing steps include handling missing values, encoding categorical variables, and normalizing numerical features. Both K-Means and DBSCAN algorithms are implemented, and their performance is evaluated using metrics such as silhouette score, Davies-Bouldin index, and Calinski-Harabasz score. The results indicate that K-Means achieved a silhouette score of 0.546, a Davies-Bouldin index of 0.655, and a Calinski-Harabasz score of 552.9. In contrast, DBSCAN achieved a higher silhouette score of 0.680, a Davies-Bouldin index of 1.344, and a Calinski-Harabasz score of 1123.9. These findings suggest that while DBSCAN performs better in terms of silhouette score, indicating more distinctly separated clusters, its higher Davies-Bouldin index reflects fewer compact clusters. The discussion highlights that K-Means is suitable for applications requiring clear and well-defined segments of customers, as it produces balanced cluster sizes. DBSCAN, with its strength in identifying clusters of varying densities and handling noise, is more effective in detecting niche markets and unique customer behaviors. This study's findings have significant practical implications for e-commerce businesses looking to enhance their customer segmentation strategies. In conclusion, both K-Means and DBSCAN demonstrate their respective strengths and weaknesses in clustering the e-commerce customer dataset. The choice of algorithm should be based on the specific requirements of the segmentation task. Future research could explore hybrid methods combining the strengths of both algorithms and incorporate additional data sources for a more comprehensive analysis.

Comparative Analysis of Sentiment Classification Techniques on Flipkart Product Reviews: A Study Using Logistic Regression, SVC, Random Forest, and Gradient Boosting Henderi; Siddique, Quba
Journal of Digital Market and Digital Currency Vol. 1 No. 1 (2024): Regular Issue June 2024
Publisher : Bright Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47738/jdmdc.v1i1.4

Sentiment analysis plays a crucial role in e-commerce, providing valuable insights from customer reviews on platforms like Flipkart. This study aims to compare the effectiveness of various sentiment classification techniques, specifically Logistic Regression, Support Vector Classifier (SVC), Random Forest, and Gradient Boosting. The dataset, collected from Flipkart, consists of 205,052 product reviews spanning various categories. Key data preprocessing steps included handling missing values, removing duplicates, normalizing text, and applying TF-IDF vectorization for feature extraction. We implemented and tuned the hyperparameters for each algorithm using grid search and randomized search. The data was divided into training and testing sets with an 80-20 split, and cross-validation techniques ensured robust model evaluation. The performance of each model was assessed using several metrics: accuracy, precision, recall, F1-score, and ROC-AUC. The results revealed that Logistic Regression achieved an accuracy of 0.8995, precision of 0.8773, recall of 0.8995, an F1 score of 0.8736, and a ROC AUC score of 0.9105. The SVC model showed slightly higher accuracy at 0.8997, precision of 0.8619, recall of 0.8997, and an F1 score of 0.8738. The Random Forest model, while robust, had lower accuracy (0.7953) and struggled with precision (0.6326), recall (0.7953), and an F1 score of 0.7047, but achieved a ROC AUC score of 0.9037. Gradient Boosting performed comparably to Logistic Regression with an accuracy of 0.8993, precision of 0.8512, recall of 0.8993, an F1-score of 0.8735, and a ROC AUC score of 0.9098. Comparative analysis identified SVC and Logistic Regression as top performers, balancing accuracy and computational efficiency. These findings suggest that implementing these models can significantly enhance sentiment analysis in e-commerce, improving customer insights and business strategies. Future research should explore advanced deep learning techniques and address class imbalances to further refine sentiment analysis capabilities.

Predicting Campaign ROI Using Decision Trees and Random Forests in Digital Marketing Hayadi, B Herawan; El Emary, Ibrahiem M. M.
Journal of Digital Market and Digital Currency Vol. 1 No. 1 (2024): Regular Issue June 2024
Publisher : Bright Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47738/jdmdc.v1i1.5

Digital marketing has become a cornerstone of modern business strategies, leveraging various channels and technologies to promote products and services. Measuring the Return on Investment (ROI) is crucial in evaluating the effectiveness of these marketing campaigns. This study aims to predict the ROI of digital marketing campaigns using two prominent machine learning algorithms: Decision Trees and Random Forests. The primary objective of this research is to compare the performance of Decision Trees and Random Forests in predicting the ROI of digital marketing campaigns. The study focuses on evaluating the accuracy, precision, and robustness of these models, and identifying the key features that influence ROI. The dataset used in this study comprises 200,000 rows and 16 columns, detailing various aspects of digital marketing campaigns, including campaign type, target audience, duration, and channels used. Initial Exploratory Data Analysis (EDA) identified no missing values or duplicates, ensuring a clean dataset for modeling. Data preprocessing involved feature engineering and encoding categorical variables. The models were trained and evaluated using an 80-20 split for training and testing, with cross-validation applied to ensure robustness. The Decision Tree model achieved a Mean Squared Error (MSE) of 1.0896, a Root Mean Squared Error (RMSE) of 1.0439, a Mean Absolute Error (MAE) of 0.8958, and an R2 value of -0.0781. In contrast, the Random Forest model showed superior performance with an MSE of 1.0143, an RMSE of 1.0071, an MAE of 0.8755, and an R2 value of -0.0035. Cross-validation for the Random Forest model yielded a CV MSE of 1.0035, a CV RMSE of 1.0018, and a CV R2 of -0.0039, reinforcing its robustness and accuracy. The Random Forest model's superior performance is attributed to its ability to handle complex interactions between features and its robustness against overfitting. Key predictors such as Conversion_Rate, Acquisition_Cost, and Engagement_Score were identified as significant factors influencing ROI. The study discusses the practical implications of these findings for optimizing digital marketing strategies, acknowledging the limitations of data quality and model assumptions, and suggesting directions for future research, including the integration of additional data sources and exploration of advanced machine learning techniques. This study highlights the potential of machine learning models, particularly Random Forests, in predicting the ROI of digital marketing campaigns. The findings provide valuable insights for marketers to enhance their strategies and optimize budget allocations, emphasizing the importance of predictive analytics in achieving marketing success. Future work should focus on improving model accuracy and exploring new techniques to further advance the field of marketing analytics.

Comparative Analysis of Ensemble Learning Techniques for Purchase Prediction in Digital Promotion through Social Network Advertising Hananto, Andhika Rafi; Srinivasan, Bhavana
Journal of Digital Market and Digital Currency Vol. 1 No. 2 (2024): Regular Issue September
Publisher : Bright Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47738/jdmdc.v1i2.7

This study conducts a comprehensive comparative analysis of ensemble learning techniques for predicting user purchases in social network advertising. The ensemble methods evaluated include Random Forest, Gradient Boosting Machines (GBM), AdaBoost, and Bagging. The dataset, consisting of 7,000 records of user interactions with social network advertisements, was preprocessed to handle missing values, encode categorical variables, and standardize numerical features. Performance metrics such as accuracy, precision, recall, F1 score, and ROC AUC score were used to evaluate each model. The Random Forest model achieved an accuracy of 0.875, precision of 0.821, recall of 0.821, F1 score of 0.821, and ROC AUC score of 0.948. The GBM model also performed well, with an accuracy of 0.875, precision of 0.846, recall of 0.786, F1 score of 0.815, and ROC AUC score of 0.948. The AdaBoost model showed the highest performance, with an accuracy of 0.9, precision of 0.917, recall of 0.786, F1 score of 0.846, and ROC AUC score of 0.969. The Bagging model achieved an accuracy of 0.875, precision of 0.821, recall of 0.821, F1 score of 0.821, and ROC AUC score of 0.939. Feature importance analysis revealed that Age and Estimated Salary were the most significant predictors across all models. Hyperparameter tuning was crucial in optimizing each model's performance, ensuring they were neither too simple nor too complex. The study's findings underscore the effectiveness of ensemble learning techniques in social network advertising and provide valuable insights for marketers. Future research could explore larger and more diverse datasets, other ensemble methods, and the computational efficiency of these models. This research contributes to predictive analytics in marketing, enhancing the accuracy and effectiveness of advertising strategies.

Modeling the Impact of Holidays and Events on Retail Demand Forecasting in Online Marketing Campaigns using Intervention Analysis Saputra, Jeffri Prayitno Bangkit; Kumar, Aayush
Journal of Digital Market and Digital Currency Vol. 1 No. 2 (2024): Regular Issue September
Publisher : Bright Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47738/jdmdc.v1i2.9

This study explores the impact of holidays and events on retail demand forecasting using intervention analysis within a SARIMAX model framework. Retail demand forecasting is critical for inventory management and supply chain optimization. Traditional forecasting models often struggle to account for irregular events like holidays, leading to inaccuracies. This study aims to address these limitations by incorporating holidays and events as exogenous variables in the forecasting model. The dataset, consisting of retail sales records across multiple product categories, was preprocessed to handle missing values and standardize date formats. Binary indicators for state holidays and school holidays were created, along with temporal features like the day of the week and hour of the day. The stationarity of the time series was confirmed using the Augmented Dickey-Fuller (ADF) test, with a statistic of -48.67066391486136 and a p-value of 0.0. The SARIMAX model (1, 1, 1)x(1, 1, 1, 24) was developed and evaluated. The model achieved an Akaike Information Criterion (AIC) of 363321.861 and a Bayesian Information Criterion (BIC) of 363375.269. Key coefficients included the state holiday variable at 0 (p-value: 1.000000) and the school holiday variable at 165.2158 (p-value: 0.919689), though neither were statistically significant. Diagnostic checks revealed significant non-normality and heteroscedasticity in the residuals. Forecasting accuracy was assessed using Mean Absolute Error (MAE: 8057.069376036054) and Mean Squared Error (MSE: 809008799.3517022). The Mean Absolute Percentage Error (MAPE) was not computable due to division by zero. Visualizations comparing forecasted versus actual demand highlighted the model’s strengths in capturing general trends and seasonal patterns but indicated challenges in accurately predicting demand during holidays and events. The study underscores the importance of incorporating holidays and events into demand forecasting models and suggests further refinement and the inclusion of additional variables for improved accuracy. Future research should explore alternative modeling approaches and validate findings across multiple datasets to enhance the generalizability and robustness of the forecasting tools.

Temporal Patterns in User Conversions: Investigating the Impact of Ad Scheduling in Digital Marketing Pratama, Satrya Fajri; Sugianto, Dwi
Journal of Digital Market and Digital Currency Vol. 1 No. 2 (2024): Regular Issue September
Publisher : Bright Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47738/jdmdc.v1i2.10

This study explores the impact of ad scheduling on user conversions by analyzing temporal patterns in user behavior. In the increasingly competitive landscape of digital marketing, optimizing the timing of ad placements is critical for maximizing user engagement and conversion rates. Utilizing a comprehensive dataset from Kaggle, which includes variables such as user ID, ad exposure details, and conversion outcomes, we employed both time series analysis and survival analysis to uncover insights into how different ad scheduling strategies affect conversion rates. The ARIMA model, used for time series analysis, provided reasonable predictive accuracy with a Mean Absolute Error (MAE) of 389.92, Root Mean Squared Error (RMSE) of 463.97, and Mean Absolute Percentage Error (MAPE) of 2.26%. This model effectively identified specific hours and days with higher likelihoods of conversion, particularly during evenings and weekends. On the other hand, the Cox Proportional Hazards model, used for survival analysis, demonstrated superior performance with a concordance index of 0.97, indicating its exceptional ability to predict the timing of user conversions based on various covariates such as the number of ads seen and the specific hours of exposure. The findings suggest that strategic ad scheduling, tailored to align with user temporal behavior, can significantly enhance marketing effectiveness by targeting users during peak conversion periods. These insights offer practical implications for digital marketers aiming to refine their ad delivery strategies to achieve higher conversion rates and improve return on investment.

Customer Segmentation and Targeted Retail Pricing in Digital Advertising using Gaussian Mixture Models for Maximizing Gross Income Hariguna, Taqwa; Chen, Shih Chih
Journal of Digital Market and Digital Currency Vol. 1 No. 2 (2024): Regular Issue September
Publisher : Bright Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47738/jdmdc.v1i2.11

This study investigates the application of Gaussian Mixture Models (GMM) for customer segmentation and targeted pricing strategies in the retail industry to maximize gross income. Using a dataset of 1000 transaction records, the analysis focused on attributes such as unit price, quantity, total amount, and payment methods. The dataset was preprocessed to handle missing values, encode categorical features, and scale numerical features. The optimal number of components for the GMM was determined using the Bayesian Information Criterion (BIC), resulting in the selection of 10 clusters. Model training was conducted using the Expectation-Maximization (EM) algorithm, achieving convergence after 18 iterations. Customer segments were identified and analyzed based on their purchasing behaviors and demographic traits. For instance, Segment 0 preferred bulk purchases of lower-priced items, while Segment 1 favored higher-priced items in smaller quantities, resulting in a higher average purchase value of 2274.19. Conversely, Segment 2 showed a high frequency of returns, indicated by a negative average purchase value of -2608.40. Targeted pricing strategies were developed for each segment, aiming to maximize gross income. The effectiveness of the segmentation and pricing strategies was evaluated using metrics such as the silhouette score, with training and testing scores of 0.175 and 0.015 respectively, highlighting areas for improvement in clustering quality. This study underscores the potential of GMM in uncovering distinct customer segments and tailoring pricing strategies to enhance profitability. Future research should explore alternative clustering techniques and extend the analysis to other retail domains and larger datasets to validate and improve the findings. The practical implications for retail businesses include the need for iterative testing and refinement of pricing strategies based on customer segmentation to achieve sustainable growth and customer satisfaction.

Evaluating Behavioral Intention and Financial Stability in Cryptocurrency Exchange App: Analyzing System Quality, Perceived Trust, and Digital Currency Yadulla, Akhila Reddy; Nadella, Geeta Sandeep; Maturi, Mohan Harish; Gonaygunta, Hari
Journal of Digital Market and Digital Currency Vol. 1 No. 2 (2024): Regular Issue September
Publisher : Bright Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47738/jdmdc.v1i2.12

This study evaluates the factors influencing financial stability (FS) and behavioral intention (BI) in a cryptocurrency exchange app, explicitly focusing on system quality (SQ), perceived trust (PT), and digital currency (DC) within the Indonesian context. Utilizing structural equation modeling (SEM) with SmartPLS, the research analyzed data from 345 respondents who are active users of the cryptocurrency exchange app. The results confirmed that SQ significantly enhances PT (β = 0.832, t = 27.216, p < 0.001) and BI (β = 0.718, t = 12.675, p < 0.001). Additionally, DC positively impacts FS (β = 0.578, t = 8.177, p < 0.001), while PT influences both FS (β = 0.391, t = 5.478, p < 0.001) and BI (β = 0.198, t = 3.490, p = 0.001). These findings validate all five proposed hypotheses, highlighting the critical role of SQ and PT in driving FS and user engagement in cryptocurrency exchange apps. The study's measurement model demonstrated good reliability and validity, with Cronbach's alpha values exceeding 0.7 for all constructs: SQ (0.891), PT (0.812), DC (0.767), FS (0.819), and BI (0.745). Composite reliability values were also high, ranging from 0.855 to 0.933. Average Variance Extracted (AVE) values indicated good convergent validity, with SQ (0.822), PT (0.727), DC (0.689), FS (0.743), and BI (0.663). Discriminant validity was confirmed using the Fornell-Larcker criterion. The structural model's fit indices, including an SRMR of 0.045 and an NFI of 0.914, demonstrated a good model fit. The R² values for BI (0.791), FS (0.873), and PT (0.693) indicated substantial explanatory power. Despite its contributions, this study has limitations, including its focus on a single cryptocurrency exchange app in Indonesia, which may affect the generalizability of the findings. Future research should expand the sample to include multiple apps and geographical contexts. Additionally, incorporating other relevant factors, such as user experience and regulatory compliance, could provide a more comprehensive understanding of FS in digital financial services. This research underscores the importance of SQ and PT in achieving long-term success and sustainability in the rapidly evolving digital finance landscape.

Home Page

OAI Link

Editorial Team

Contact

Reviewer

Google Scholar

Contact Name
Andhika Rafi Hananto

Contact Email
andhikarh90@gmail.com

Phone
+6282314736799

Journal Mail Official
support@jdmdc.com

Editorial Address
Graha Permata Estate, Jl. HM Bahrun Blok H9, Sokayasa, Berkoh, Kec. Purwokerto Tim., Kabupaten Banyumas, Jawa Tengah 53146

Location
Kab. banyumas,

Jawa tengah

INDONESIA

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Home Page

OAI Link

Editorial Team

Contact

Reviewer

Google Scholar

Contact Name Andhika Rafi Hananto

Contact Email andhikarh90@gmail.com

Phone +6282314736799

Journal Mail Official support@jdmdc.com

Editorial Address Graha Permata Estate, Jl. HM Bahrun Blok H9, Sokayasa, Berkoh, Kec. Purwokerto Tim., Kabupaten Banyumas, Jawa Tengah 53146

Location Kab. banyumas, Jawa tengah INDONESIA

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Contact Name
Andhika Rafi Hananto

Contact Email
andhikarh90@gmail.com

Phone
+6282314736799

Journal Mail Official
support@jdmdc.com

Editorial Address
Graha Permata Estate, Jl. HM Bahrun Blok H9, Sokayasa, Berkoh, Kec. Purwokerto Tim., Kabupaten Banyumas, Jawa Tengah 53146

Location
Kab. banyumas,

Jawa tengah

INDONESIA