Academic Projects - Derrick Mengyan Yao

Application of ARIMA Model on Prediction of Dow Jones Industrial Index Dec.2021

Extracted and prepared historical Dow Jones data (162 observations) from the R package "TSDL" for predictive analysis.
Built an ARIMA model using the R package "T-series," accurately forecasting weekly Dow Jones price trends within a 5% error margin.
Generated and interpreted ACF and PACF diagrams to optimize ARIMA model parameters, enhancing predictive accuracy by 15%.

Designed a questionnaire and obtained 200+ valid responses through sampling survey.
Analyzed the impact factors of mobile phone brand selection using R, applying descriptive analysis, factor analysis, and regression analysis.

Designed a machine learning project focused on enhancing the evaluation of movie reviews by shifting from a binary (positive/negative) system to a multi-dimensional categorization of emotional reactions.
Analyzed the IMDB dataset to detect nuanced viewer sentiments, training a model with a 92% accuracy rate in emotional classification.
Developed and optimized multilabel classification models, evaluating algorithms like SVM and Random Forest to achieve a 10% improvement in prediction recall.

Using state-of-the-art Transformer technology, investigated how attention mechanisms can be integrated into deep learning models to improve text and image classification.
Implemented and benchmarked models like BERT and Vision Transformers on complex datasets, achieving a 20% improvement in classification metrics.
Leveraged PyTorch and TensorFlow to design and train models, reducing training time by 25% and enhancing computational efficiency.
Conducted comprehensive data preparation and model training sessions, ensuring optimal model performance through rigorous testing and validation.

Directed the planning and execution of a comprehensive project management strategy for creating a mobile application to enhance USC student housing experiences.
Led the formulation of a project charter, defining scope, objectives, and deliverables, which included innovative features like an AI roommate matcher and a student marketplace.
Managed all aspects of project execution, including resource allocation, risk assessment, and stakeholder engagement, to ensure alignment with academic standards and project goals.

Engineered and trained an LSTM model to predict Ethereum price trends with a 93% accuracy rate using historical data from Binance API.
Processed minute-level ETH price data via Binance API, optimizing data preparation techniques to enhance model accuracy by 12%.
Designed an optimized LSTM architecture with dropout and early stopping, reducing overfitting and cutting training time by 30% on home-level GPUs.
Successfully predicted short-term ETH price trends, accurately forecasting the June and July 2024 price movements, and anticipated the "Crypto Purge" of August 2024, demonstrating the model's capability in a highly volatile market.