Seybold Report ISSN: 1533-9211
Ritesh Kumar, Dr. Narendra Sharma
Vol 18, No 11 ( 2023 ) | Licensing: CC 4.0 | Pg no: 466-487 | Published on: 30-11-2023
Abstract
This research presents a analysis of various forecasting models for predicting stock prices, total revenue, & operating profits of seven major technology stocksfrom the glove: (AAPL) Apple, (MSFT) Microsoft, Amazon (AMZN), Meta (META), (GOOGL) Alphabet, TCS (Tata Counsultancey Servises) and Infosys . The methodology involved six key steps: data collection from the yfinance library covering January 1, 2020, to January 1, 2023; feature engineering to create lagged variables; an 80/20 train/test split; model selection including naïve bayesian regression, KNN (k-nearest neighbors), ANN (artificial neural networks), simple moving average (SMA), & the exponential moving average (EMA); & performance evaluation through metrics such ad this like mean error (MAE), (MSE) mean squared error, & the mean absolute percentage error (MAPE). Results indicate that for stock price predictions, Naïve Bayesian Regression consistently outperformed ANN across all metrics for AAPL (MAE = 2.73), while for MSFT, ANN showed higher errors (MAE = 5.24). In total revenue predictions, traditional methods like SMA & EMA yielded zero error metrics, highlighting their robustness compared to more complex models such as ANN, which produced MAE values as high as 43.08 billion for AAPL. Operating profit predictions revealed that while traditional models maintained low MAE (SMA: 9.71 billion, EMA: 4.96 billion for AAPL), ANN exhibited MAE of 1.02e+11, indicating substantial overfitting. the findings emphasize that simpler models often outperform advanced techniques in forecasting financial metrics, underscoring the critical importance of model selection tailored to specific predictive goals.
Keywords:
Stock price prediction, revenue forecasting, machine learning, Naïve Bayesian Regression, Artificial Neural Networks, financial metrics.