KMITL

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

Browse

Settings

Author: search.filters.author.Boongasame, LaorSubject: search.filters.subject.deep learning

Search Results

Now showing 1 - 4 of 4
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Item,
    An Extension of Laor Weight Initialization for Deep Time-Series Forecasting: Evidence from Thai Equity Risk Prediction
    (2025-09-01)
    Petchpol, Katsamapol
    ;
    Boongasame, Laor
    This study presents a gradient-informed proxy initialization framework designed to improve training efficiency and predictive performance in deep learning models for time-series forecasting. The method extends the Laor Initialization approach by introducing backward gradient norm clustering as a selection criterion for input-layer weights, evaluated through a lightweight, architecture-agnostic proxy model. Only the numerical input layer adopts the selected initialization, while internal components retain standard schemes such as Xavier, Kaiming, or Orthogonal, maintaining compatibility and reducing overhead. The framework is evaluated on a real-world financial forecasting task: identifying high-risk equities from the Thai Market Surveillance Measure List, a domain characterized by label imbalance, non-stationarity, and limited data volume. Experiments across five architectures, including Transformer, ConvTran, and MMAGRU-FCN, show that the proposed strategy improves convergence speed and classification accuracy, particularly in deeper and hybrid models. Results in recurrent-based models are competitive but less pronounced. These findings support the method’s practical utility and generalizability for forecasting tasks under real-world constraints.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Item,
    Laor Initialization: A New Weight Initialization Method for the Backpropagation of Deep Learning
    (2025-07-01)
    Boongasame, Laor
    ;
    Muangprathub, Jirapond
    ;
    Thammarak, Karanrat
    This paper presents Laor Initialization, an innovative weight initialization technique for deep neural networks that utilizes forward-pass error feedback in conjunction with k-means clustering to optimize the initial weights. In contrast to traditional methods, Laor adopts a data-driven approach that enhances convergence’s stability and efficiency. The method was assessed using various datasets, including a gold price time series, MNIST, and CIFAR-10 across the CNN and LSTM architectures. The results indicate that the Laor Initialization achieved the lowest K-fold cross-validation RMSE (0.00686), surpassing Xavier, He, and Random. Laor demonstrated a high convergence success (final RMSE = 0.00822) and the narrowest interquartile range (IQR), indicating superior stability. Gradient analysis confirmed Laor’s robustness, achieving the lowest coefficients of variation (CV = 0.2230 for MNIST, 0.3448 for CIFAR-10, and 0.5997 for gold price) with zero vanishing layers in the CNNs. Laor achieved a 24% reduction in CPU training time for the Gold price data and the fastest runtime on MNIST (340.69 s), while maintaining efficiency on CIFAR-10 (317.30 s). It performed optimally with a batch size of 32 and a learning rate between 0.001 and 0.01. These findings establish Laor as a robust alternative to conventional methods, suitable for moderately deep architectures. Future research should focus on dynamic variance scaling and adaptive clustering.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Item,
    Thai Morning Glory Price Forecasting Using Deep Learning
    (2025-01-01)
    Waeodi, Kanokwan
    ;
    Boongasame, Laor
    ;
    Thammarak, Karanrat
    This study established advanced machine-learning-driven forecasting models to enhance the accuracy of price predictions for Thai morning glory, a widely consumed leafy green vegetable. The models were trained using historical price, weather, and rainfall data using time-series forecasting methods, specifically LSTM and CNN. The findings indicate that stepwise feature selection minimizes prediction errors and improves MSE, RMSE, MAPE, and MAE. Preliminary experiments revealed that the LSTM model with feature selection outperformed the other models, particularly in feature selection. Employing standard hyperparameters of 100 epochs, 32 batches, and five windows, the model demonstrated superior performance with a lower MSE (0.0010), RMSE (0.0274), MAPE (3.7803), and MAE (0.0158) than the CNN model. Statistical hypothesis testing revealed significant variations between the LSTM and CNN models, with feature selection p-values below 0.05. These results indicate that LSTM with feature selection models optimized through refined hyperparameters leads to more accurate Thai morning glory price forecasting, providing valuable insights for stakeholders in their decision-making processes. Additionally, this study can forecast prices for 5, 7, 14, and 21 days in advance based on different Window_len values, addressing various planning needs. The 5- and 7-day forecasts support short-term decision-making, such as scheduling harvest cycles and weekly market planning, whereas the 14-day forecast assists farmers in optimizing planting schedules and logistics. Furthermore, the 21-day forecast is beneficial for medium-term market planning, including negotiating forward contracts and adjusting distribution strategies to maximize profitability.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Item,
    Enhancing Predictive Capabilities for Identifying At-Risk Stocks Using Multivariate Time-Series Classification: A Case Study of the Thai Stock Market
    (2025-01-01)
    Petchpol, Katsamapol
    ;
    Boongasame, Laor
    This study proposes a multivariate time-series classification approach using deep learning to predict stocks likely to be flagged by the Market Surveillance Measure List in the Thai stock market. Formulated as a binary classification problem, the model distinguishes At-Risk and Normal stocks based on two primary datasets: End-of-Day stock prices and Market Surveillance Measure List records, incorporating trading volumes and technical indicators. To address data imbalance, concept drift, and long-term dependencies, the framework integrates feature engineering, cost-sensitive learning, and rolling window training. Experimental results show deep learning models significantly outperform traditional baseline methods in capturing financial risk patterns. The study identifies models that effectively balance predictive accuracy with computational efficiency, with performance varying based on forecasting horizons. Despite improvements from specialized techniques, the study identifies challenges in long-term financial risk prediction. These findings support market surveillance, algorithmic trading, and portfolio risk management, with future work exploring explainable AI, adaptive learning, and alternative data sources to enhance interpretability and long-term forecasting.