Predicting the short life cycle of semiconductor products for demand forecasting using ridge regularization

In high-tech industries such as semiconductors, predicting product life cycles is a critical concern due to its substantial influence on corporate profits and losses, given the strong correlation between life cycle stages and demand forecasting. In particular, the growing diversity and complexity of products, coupled with frequent changes in sales strategies driven by environmental shifts, have considerably shortened product life cycles, thereby complicating accurate prediction. This study aims to enhance the accuracy of short life cycle predictions by employing simple statistical and regression-based models, rather than relying on complex algorithms such as deep learning. Specifically, the quarterly sales data of a semiconductor company specializing in CPU production are modeled as an S-curve by transforming the data into a cumulative distribution function (CDF), followed by the application of ridge regularization to estimate the optimal coefficients of the exponential function fitted to the S-curve. This approach enables the model to achieve relatively high predictive accuracy using a small initial dataset. Furthermore, the proposed method yields a lower root mean squared error (RMSE) than conventional diffusion models, such as the Bass and Gompertz models, which are commonly employed for short life cycle prediction, as well as advanced learning-based techniques, thereby demonstrating its effectiveness in forecasting product life cycles. The statistical significance of the results across various comparative models was also validated.