KMITL

Permanent URI for this communityhttps://dspace.kmitl.ac.th/handle/123456789/1

Browse

Search Results

Now showing 1 - 10 of 35

Some of the metrics are blocked by your
consent settings
Machine Learning Models for Multi-Horizon Classification of Bitcoin Future Price Movements
(2026-06-16)
Boonpai, Sirawat
;
Netisopakul, Ponrudee
;
Anuntachai, Anuntapat
This study presents a comprehensive framework for multi-horizon classification of Bitcoin futures price movements using machine learning. Five models of Logistic Regression, Random Forest, XGBoost, LightGBM, and CatBoost were systematically evaluated across three timeframes (4h, 12h, 1d) and four classification schemes: binary (up/down) and multi-class (up/down/stable) with thresholds of 0.5%, 1.0%, and 1.5%, totaling 60 distinct experimental configurations. Rather than pursuing a strong predictive performance, this study prioritizes a rigorous comparative analysis across multiple dimensions to identify which combinations of model, timeframe, and classification scheme are most effective for Bitcoin futures. The results demonstrate that binary classification achieves the best predictive performance, with shorter timeframes yielding significantly better results, confirming the effectiveness of technical indicators in capturing the rapid price dynamics of Bitcoin futures. Notably, CatBoost achieved the highest F1-score for binary classification, while Random Forest proved the most robust model across diverse configurations. Feature importance analysis revealed that momentum-based indicators are the dominant predictors of price direction, while volatility features play a critical role in distinguishing sideways movements from directional ones in multi-class settings. Furthermore, the study demonstrates that narrower classification thresholds (0.5%) introduce noisier class boundaries and degrade performance, whereas wider thresholds (1.0%-1.5%) yield more stable results. These findings provide actionable guidelines for algorithmic trading in Bitcoin futures and establish a reproducible benchmark for future research.
Some of the metrics are blocked by your
consent settings
A Comparative Study of Machine Learning and Deep Learning Models for Technology Stock Price Prediction Using News Sentiment and Economic Indicators
(2026-01-01)
Chaiwuttisak, Pornpimol
This study developed predictive models for the closing prices of five leading technology stocks: GOOGL, MSFT, AAPL, NVDA, and META by employing five advanced machine learning and deep learning techniques: Light Gradient Boosting Machine (LightGBM), Extreme Gradient Boosting (XGBoost), Recurrent Neural Network (RNN), Gated Recurrent Unit (GRU), and Long Short-Term Memory (LSTM). The modeling framework integrated sentiment scores derived from financial news articles specific to each stock using the VADER Sentiment Analysis tool, in conjunction with a range of macroeconomic indicators. Model performance was evaluated separately for each stock using Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE) as primary metrics. To determine whether statistically significant differences existed among the predictive performance of the models across all stocks, the Friedman test was employed, followed by the Wilcoxon signed-rank test for post-hoc pairwise comparisons. The empirical results indicated that XGBoost achieved superior predictive accuracy for MSFT and AAPL, GRU outperformed other models for NVDA and META, while RNN yielded the most accurate forecasts for GOOGL.
Some of the metrics are blocked by your
consent settings
Performance Evaluation of Imputation Techniques for Telecommunications Customer Clustering
(2026-01-01)
Sukthong, Patthama
;
Charoenporn, Pattama
Missing data significantly degrades machine learning model performance in telecommunications customer analytics, leading to unreliable customer segmentation and suboptimal business decision-making. This research systematically compares seven imputation techniques across three missing mechanisms (MCAR, MAR, MNAR) and four missing rates (5%, 10%, 20%, 30%) using the Telco Customer Churn Dataset (7,043 records). Methods evaluated include traditional approaches (mean/mode, forward fill, regression), machine learning techniques (KNN, Random Forest, MICE), and deep learning (Autoencoder). We assessed model performance using normalized MAE and RMSE, and evaluated downstream effects through clustering algorithms. Results demonstrate Random Forest imputation's superior performance with MAE of 0.1568 and RMSE of 0.2123, achieving 53.7% lower error rates compared to mean/mode imputation. Statistical analysis confirmed significant performance differences (Friedman test: χ<sup>2</sup> = 55.85, p < 0.001). Interestingly, clustering performance did not directly correlate with imputation accuracy; the Autoencoder achieved the highest silhouette score (0.1510) despite moderate reconstruction accuracy. Machine learning approaches maintained robust performance across all missing data mechanisms, whereas traditional methods degraded under MNAR conditions. These findings provide evidence-based guidelines for selecting appropriate imputation techniques in telecommunications analytics, enabling improved customer segmentation and business outcomes.
Some of the metrics are blocked by your
consent settings
A Self-Adaptive Weights for K-Means Classification Algorithm
(2025-09-01)
Chenghu, Cui
;
Thammano, Arit
This paper presents an improved K-means clustering algorithm that addresses the traditional algorithm’s sensitivity to outlier and susceptibility to local optima by introducing an adaptive weight adjustment mechanism. It employs an exponential decay function to dynamically reduce the feature weights of outlier data points, effectively suppressing outliers while preserving the structure of the normal data. The proposed method retains the computational efficiency of standard K-means. Key contributions include: (a) A novel distance-based weighting strategy that progressively reduces the influence of noisy points, mitigating the impact of outliers on clustering performance. (b) An innovative form of "local dimensionality reduction" for outlier points via weight decay, which interferes only with the feature space of noisy regions while preserving the global topological structure of clean data. Extensive experiments on three benchmark datasets Iris (4-dimensional, balanced classes), Wine (13-dimensional, correlated features), and Wisconsin Breast Cancer Diagnosis (30-dimensional, imbalanced data) demonstrate the effectiveness of the approach. Compared to standard K-means, the proposed algorithm achieves accuracy improvements of 7.47% on Iris, 13.89% on Wine, and 19% on WBCD. This adaptive strategy offers a practical and efficient solution for clustering in noisy, high-dimensional environments, without the added complexity of mixture models.
Some of the metrics are blocked by your
consent settings
Data-driven prediction of failure loads in low-cost FRP-confined reinforced concrete beams
(2025-07-01)
Talpur, Shabbir Ali
;
Thansirichaisree, Phromphat
;
Anotaipaiboon, Weerachai
;
Mohamad, Hisham
;
Zhou, Mingliang
This study investigates the application of machine learning (ML) models to predict the ultimate failure load of reinforced concrete (RC) beams confined with low-cost fiber-reinforced polymers (FRP), relatively underexplored area. A dataset of 100 samples, including beams designed to fail in flexure and shear, was compiled from literature and experimental testing. Four ML models—XGBoost, Random Forest (RF), Neural Network (NN), and Decision Tree (DT)—were evaluated using k-fold cross-validation with performance metrics such as Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and R². XGBoost outperformed the other models, achieving the highest R² of 0.96 and the lowest RMSE of 12.81, while SHAP analysis identified beam height, bottom rebar strength, and beam width as key predictors. These results highlight the effectiveness of ensemble methods for predicting failure loads in RC beams and provide insights into the most influential features affecting structural performance.
Some of the metrics are blocked by your
consent settings
Development of machine learning enhanced low-cost spectrophotometer for pesticide prediction
(2025-05-15)
Murathathunyaluk, S.
;
Jinorose, M.
;
Janpetch, K.
;
Chanthapanya, N.
;
Sombatsri, W.
Conventional analytical methods for measuring pesticide concentrations, such as chromatography, offer high accuracy but require expensive instrumentation, prompting the investigation of cost-effective alternatives like smartphone-based spectrophotometers. Despite their potential, these methods face challenges related to assembly and precision, often requiring human intervention to select appropriate images for analysis. This study presents a novel, affordable spectrophotometer designed for integration with machine learning algorithms. The device captures images of two spectral bands and employs a six-step image processing methodology to prepare images for analysis. A machine learning model trained on four algorithms with feature selection and cross-validation demonstrates high accuracy in predicting chemical concentrations of coloured solutions. The approach achieves 98.5 % accuracy for KMnO<inf>4</inf> and 96.7 % for Carbosulfan solutions, comparable to high-end spectrophotometry devices. The design eliminates the need for human intervention, reducing biased selection and result manipulation. However, concentration estimation of non-coloured compounds remains inaccurate, indicating areas for further refinement.
Some of the metrics are blocked by your
consent settings
Energy Consumption Prediction and Anomaly Detection for Boiler Feed Pump in Power Plant Using Machine Learning and Deep Learning
(2025-04-01)
Khamfoy, Polawut
;
Klomwises, Yuwadee
;
Srianomai, Sakuna
Enhancing energy efficiency and operational reliability is crucial in power plant management, particularly for high-energy-consuming machines such as boiler feed water pumps (BFPs). These pumps play a vital role in the continuous generation of steam and electricity and must operate 24/7 to maintain power production stability. This study proposes the development of predictive models based on machine learning and deep learning techniques to accurately predict energy consumption and applies best models to detect anomalous behaviors in BFPs, enabling timely and preventive interventions. A dataset comprising 43,082 hourly records over five years, with 18 critical operational features, was analyzed using preprocessing and feature engineering techniques. Various predictive models were trained and evaluated, including Multiple Linear Regression, Regularized Regressions (Ridge, Lasso, ElasticNet), Support Vector Regression (SVR), Decision Tree, Ensemble Methods (Random Forest, XGBoost, CatBoost, LightGBM), and Deep Learning Architectures (DNN, RNN, GRU, LSTM). Among these models, SVR demonstrated the highest accuracy (MSE: 13.5573, R²: 0.9838), followed closely by LightGBM. Feature importance analysis revealed that boiler feed pump discharge pressure and bearing housing vibration levels were the most influential variables in energy consumption prediction. Anomaly detection using the Interquartile Range (IQR) method classified deviations into two warning levels, enabling proactive maintenance strategies. Additionally, a Graphical User Interface (GUI) web application was developed for real-time monitoring, integrating predictive models, anomaly detection, and an automated email alert system to assist operators in responding to abnormal energy consumption events promptly. These results highlight the potential of predictive analytics and real-time monitoring in optimizing power plant operations, providing a foundation for extending predictive capabilities to other critical energy-intensive systems.
Some of the metrics are blocked by your
consent settings
Classifying Lexical Variations in Thai Social Media: A Machine Learning and LLMs Approach
(2025-01-01)
Chirawat, Pimmada
;
Leenoi, Dhanon
;
Takhom, Akkharawoot
;
Boonkwan, Prachya
;
Utasri, Tharathon
Social media platforms like Facebook, X (formerly Twitter), and TikTok produce large amounts of user content, offering valuable data for Social Media Analytics (SMA). However, analyzing Thai social media presents challenges due to lexical variations, such as spelling changes, semantic shifts, and the creation of new words (lexicogenesis). These features, along with the lack of spaces between words in Thai script, complicate traditional text analysis methods. This study addresses these challenges by using a combination of machine learning (ML) models and large language models (LLMs). We employ Bidirectional Long Short-Term Memory (Bi-LSTM) networks, pretrained models like BERT-base-Thai, and advanced LLMs such as LLaMA-3, Mistral, Gemma, and Phi to classify and analyze Thai social media content. Our research enhances the understanding of lexical variations and improves the processing of Thai language in social media, contributing to the development of more effective natural language processing tools.
Some of the metrics are blocked by your
consent settings
Recognition of Marijuana Plant Leaf Diseases Based on Deep Learning
(2025-01-01)
Kong, Qingye
;
Tooprakai, Siraphop
Plant pests and diseases are a common problem in agricultural production, and if not handled properly, they can seriously affect crop yield and quality. The era of the Internet of Things has arrived, and modern monitoring methods have also been applied to the growth monitoring of some crops, achieving good results. However, the cultivation of marijuana still relies mainly on traditional manual monitoring and modern methods have not yet been widely used. At the same time, the prevention of diseases and pests in marijuana is the focus, and if problems are not detected early, the losses can often be severe. This article proposes using machine learning to screen for abnormal leaves and confirm whether the leaves are healthy. After repeated training, the system can compare and classify different images of marijuana leaves and identify abnormal parts. The system is designed based on Python. The test results indicate that this technology can be applied to distinguish marijuana leaves to ensure early detection of diseases and pests, and to minimize agricultural losses caused by untimely remedial measures.
Some of the metrics are blocked by your
consent settings
Predicting Faulty Production Lines Causing Head Gimbal Assembly Damage from Electrostatic Discharge Using Machine Learning
(2025-01-01)
Bilgasun, Nuttanon
;
Thanasopon, Bundit
This research was conducted to analyze abnormal production lines leading to Head Gimbal Assembly (HGA) damage, mainly caused by electrostatic discharge. HGAs are critical component in hard disk. The HGA was constantly developed to produce higher storage capacity products to delight customer demands. Newer product design HGAs are highly sensitive to electrostatic discharge. This results in more damage and higher scrap costs. The main goal of this research is to leverage artificial intelligence technology to predict abnormal production lines that cause electrostatic damage in the assembly process of HGA. The research aims to develop timely solutions to reduce damage during production. This proactive approach is essential in lowering production costs and ensuring the HGAs have a longer lifespan and increased reliability.

KMITL

Browse

Filters

Settings

Sort By

Results per page

Search Results