Irwansyah
Teknik Informatika, Universitas Muhammadiyah Prof. Dr. Hamka, Jakarta

Published : 7 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 7 Documents
Search

Klasifikasi Metode Naïve Bayes pada Ulasan Pengguna Aplikasi Dazzcam untuk Pengeditan Foto Vintage di App Store Salsa Dwi Agistina; Irwansyah; Agus Budiyantara; Faldy Irwiensyah
DIGINTEL-AI : DIGital INnovation and inTELligence – AI Vol. 1 No. 2 (2026): April
Publisher : PT Ajira Karya Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.66217/digintel-ai.v1i2.9

Abstract

The rapid growth of mobile applications has increased the importance of user-generated reviews as a source of information for evaluating application quality and user satisfaction. Dazzcam, a photo editing application known for its vintage-style filters, has gained significant popularity among iOS users. This study aims to classify user reviews from the App Store into positive and negative sentiment categories using the Naïve Bayes algorithm and to evaluate the performance of the model. A total of 911 reviews were collected and divided into training and testing datasets with a ratio of 80:20. The research methodology includes data preprocessing, feature extraction using TF-IDF, and classification using Naïve Bayes, followed by evaluation with a confusion matrix. The results show that 712 reviews were classified as positive and 199 as negative, with an accuracy of 79.78%, precision of 79.89%, recall of 79.78%, and F1-score of 79.53%. These findings indicate that the Naïve Bayes algorithm demonstrates good performance and can be effectively utilized for sentiment analysis of application reviews.
Penerapan Algoritma XGBoost dalam Klasifikasi Jumlah Korban Kecelakaan Kereta Api di Indonesia Selphia Nur Azzahra; Irwansyah; Tupan Tri M; Firman Noor Hasan
DIGINTEL-AI : DIGital INnovation and inTELligence – AI Vol. 1 No. 2 (2026): April
Publisher : PT Ajira Karya Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.66217/digintel-ai.v1i2.10

Abstract

This study aims to classify the number of vehicle accident casualties caused by railway accidents in Indonesia into low, medium, and high-risk categories using the XGBoost algorithm, as well as to evaluate the model performance based on accuracy, precision, and recall metrics. The employed methodology is CRISP-DM, consisting of stages such as business understanding, data understanding, data preparation, modeling, evaluation, and deployment stages. The dataset was obtained from official reports of the National Transportation Safety Committee (KNKT) and online news articles from 1991 to early 2025, resulting in 112 valid records after preprocessing, including data labeling, transformation of nominal attributes, and conversion of date data into numerical form. The classification process was carried out using RapidMiner. The results show that the XGBoost model achieved an accuracy of 88.39%, with the highest precision and recall values in the low-risk class (0.91 and 0.94) and high-risk class (0.88 and 0.87), while the performance for the medium-risk class remains relatively low (precision 0.75 and recall 0.68), indicating potential data imbalance or insufficient discriminative features. Based on these findings, it can be concluded that the XGBoost algorithm is effective in classifying railway accident risk levels; however, improvements in data quality and feature selection are still needed to achieve more optimal performance.
Perbandingan Kinerja Algoritma K-Nearest Neighbor dan Decision Tree dalam Analisis Sentimen Ulasan Aplikasi DANA pada Google Play Store Khofifah Dwi Fany; Irwansyah; Moh Shidqon; Nunik Pratiwi
DIGINTEL-AI : DIGital INnovation and inTELligence – AI Vol. 1 No. 2 (2026): April
Publisher : PT Ajira Karya Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.66217/digintel-ai.v1i2.11

Abstract

The rapid growth of digital wallet applications such as DANA has raised concerns regarding the quality of services provided to users. One effective approach to evaluate service quality is through sentiment analysis of user reviews on the Google Play Store platform. However, the large volume of available review data makes manual analysis inefficient. This study aims to identify the most optimal classification algorithm for sentiment analysis of DANA application reviews by comparing the performance of the K-Nearest Neighbor (K-NN) and Decision Tree algorithms. The dataset consists of 723 reviews obtained from Kaggle, divided into 578 training data and 145 testing data. The reviews are classified into three sentiment categories: positive, negative, and neutral. The research process includes data collection, filtering, preprocessing (case folding, tokenizing, stopword removal, and token length filtering), TF-IDF weighting, implementation of classification algorithms, and evaluation using a Confusion Matrix. The results show that the K-NN algorithm achieves an accuracy of 53.10%, precision of 90.32%, and recall of 41.79%, while the Decision Tree algorithm yields a higher recall but lower accuracy and precision. Based on the comparison of these evaluation metrics, the K-NN algorithm is recommended as the more optimal method, as it provides a better balance between prediction accuracy and error rate compared to the Decision Tree.
Implementasi Algoritma Long Short-Term Memory untuk Memprediksi Harga Mata Uang Kripto Litecoin Muhamad Jafar Rahadian; Irwansyah; Agus Indra P; Estu Sinduningrum
DIGINTEL-AI : DIGital INnovation and inTELligence – AI Vol. 1 No. 2 (2026): April
Publisher : PT Ajira Karya Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.66217/digintel-ai.v1i2.12

Abstract

This study aims to evaluate the effectiveness of the Long Short-Term Memory (LSTM) algorithm in predicting the price of the Litecoin cryptocurrency. The dataset used consists of historical Litecoin price data against USD obtained from Yahoo Finance. Considering the high volatility of the cryptocurrency market, accurate price prediction is essential to assist investors in minimizing risks and maximizing potential returns. The LSTM method was selected due to its capability to model time-series data and capture long-term dependencies. The results show that the LSTM model is able to generate accurate predictions, achieving a Root Mean Square Error (RMSE) of 3.72% and a coefficient of determination (R²) of 91.38%. These findings indicate that the LSTM algorithm has strong potential for cryptocurrency price prediction, particularly for Litecoin.
Perbandingan Naive Bayes Classifier dan SVM untuk Analisis Sentimen Desain Seragam Atlet Indonesia pada Media Sosial X di Olimpiade Paris 2024 Azizah Salma Nida; Irwansyah; Ade Davy Wiranata; Zuhri Halim
DIGINTEL-AI : DIGital INnovation and inTELligence – AI Vol. 1 No. 2 (2026): April
Publisher : PT Ajira Karya Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.66217/digintel-ai.v1i2.13

Abstract

The Olympics is an international sporting event held every four years and serves as a platform for countries to showcase their athletic capabilities and national identity. One aspect that attracts public attention is the design of athletes' uniforms, which not only have aesthetic value but also support athletic performance. Differences in public perception of these designs generate various opinions expressed on social media X. This study aims to analyze public sentiment toward the design of Indonesian athletes' uniforms at the Paris 2024 Olympics on social media X and to compare the performance of Naive Bayes Classifier and Support Vector Machine algorithms. The dataset consists of textual data collected from social media X and processed through preprocessing stages and split into training and testing data with an 80:20 ratio. The results show that there are 1,014 positive and 728 negative sentiments. Model evaluation indicates that the Naive Bayes Classifier achieved an accuracy of 80.5%, while the Support Vector Machine achieved 94.2%, outperforming the former. These findings demonstrate that the Support Vector Machine is more effective than the Naive Bayes Classifier for sentiment analysis of social media text data related to the design of Indonesian athletes' uniforms at the Paris 2024 Olympics.
Klasifikasi Konten AI-Generated dan Non-AI pada Teks Hashtag dan Komentar TikTok Menggunakan Logistic Regression Khalifah Safa Rizky Khalifah; Irwansyah; Tupan Tri M
DIGINTEL-AI : DIGital INnovation and inTELligence – AI Vol. 2 No. 1 (2026): October
Publisher : PT Ajira Karya Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.66217/digintel-ai.v2i1.21

Abstract

The widespread distribution of AI-generated content on TikTok raises concerns about disinformation and the authenticity of online interaction. Most previous studies emphasise visual features, whereas textual features such as hashtags and user comments remain underexplored. This study aims to classify AI-generated and non-AI content on TikTok using the text of hashtags and user comments with the Logistic Regression method. Data were collected through web scraping with Apify and, after cleaning, 1,008 text records were obtained. Preprocessing consisted of case folding, removal of URLs, mentions, hashtags and non-alphabetic characters, tokenisation, stop-word filtering and stemming. The text was converted into numerical features using Term Frequency–Inverse Document Frequency (TF-IDF), and the data were split 70:30 into 705 training and 303 testing records. The model was built and evaluated in RapidMiner Studio using a confusion matrix. It achieved an overall accuracy of 72.28%. For the non-AI class, precision, recall and F1-score were 74.45%, 86.67% and 80.09%, whereas for the AI-generated class they were 65.79%, 46.30% and 54.35%. The low recall of the AI-generated class was caused by 58 false negatives. These findings indicate that Logistic Regression with TF-IDF can serve as a lightweight baseline for text-based detection of AI-generated content on TikTok, but class imbalance and the ambiguity of short, informal text still limit its sensitivity.
Klasifikasi Kualitas Udara Berdasarkan Parameter Lingkungan Menggunakan Algoritma Decision Tree C4.5 Galang Putra Wicaksana Galang; Irwansyah; Ahmad Taufik; Faldy Irwiensyah
DIGINTEL-AI : DIGital INnovation and inTELligence – AI Vol. 2 No. 1 (2026): October
Publisher : PT Ajira Karya Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.66217/digintel-ai.v2i1.22

Abstract

Air quality directly affects public health. In Indonesia it is communicated through the Air Pollution Standard Index (ISPU), which is calculated from six pollutants: PM2.5, PM10, SO2, CO, O3, and NO2. This study aims to build an accurate and interpretable model that classifies ISPU categories using the C4.5 Decision Tree algorithm. The secondary dataset from Kaggle contains 1,416 air quality monitoring records from DKI Jakarta for 2021–2025. Preprocessing consisted of median and mode imputation for missing values, removal of one unlabeled record, merging the very small “Very Unhealthy” class (4 records) into “Unhealthy”, and label encoding, which resulted in 1,415 records in three classes (Good, Moderate, Unhealthy). The data were split 70:30 with stratification into 990 training and 425 testing records. The tree was built from entropy and information gain, pruned from 39 to 19 nodes (depth from nine to five levels), and converted into ten IF-THEN rules. PM2.5 had the highest information gain (0.8717) and became the root node, with a feature importance of 0.9048. The model correctly classified 416 of 425 test records, an accuracy of 97.88%. The F1-scores for the Unhealthy, Moderate, and Good classes were 99.35%, 98.27%, and 69.57%, respectively; the lower score for Good reflects only 11 test records. SMOTE and ADASYN did not improve accuracy (95.06% and 95.53%). The results show that C4.5 can classify ISPU categories accurately using transparent rules.