Intern(Chow Jun Hoe)

 30/5.2025 I held a meeting with my manager today and gave a detailed report on the sales data analysis I had completed. I explained and responded to the manager's questions, including data inconsistencies and script usage. After the meeting, I wrote a summary article of about 400 words. I then continued to study the SARIMA model and gain a deeper understanding of how seasonal factors affect sales changes.

 29/5/2025 After successfully completing the script writing and understanding of the principles of the ARIMA model, we officially entered the learning stage of the SARIMA (Seasonal ARIMA) model. SARIMA adds seasonal components to ARIMA and is used to process time series data with obvious cyclical fluctuations. It is especially suitable for seasonal sales fluctuations commonly seen in retail, supply chain and other scenarios.

 28/5/2025 After deeply understanding the importance of ACF and PACF for ARIMA model settings, continue to expand learning other key statistical concepts and test methods to improve the ability of model evaluation and interpretation, including p value, cofidence interval, Ljung-box, etc.

 27/5/2025 Continue to study the ARIMA model in depth, focusing on learning how to use ACF (autocorrelation function) and PACF (partial autocorrelation function) graphs to judge the parameter settings (p, d, q) in the model. To ensure the quality of the model, start learning how to use AIC (Akaike Information Criterion) and BIC (Bayesian Information Criterion) to evaluate the pros and cons of the model under different parameter combinations and select the optimal model structure.

 26/5/2025 I continued to advance and complete the newly assigned analytical tasks, deeply processing and mining relevant data. I then attended internal company meetings to further understand the company's operating model and was assigned a new task, which was to gain a deeper understanding of how the company's KPIs (key performance indicators) were calculated.

 23/5/2025 After organizing and perfecting all the report contents, I successfully reported to the relevant supervisor. After the report, I accepted a new analysis task and began to enter a more in-depth data mining phase. I continued to use Python to gain a more comprehensive insight into sales.

 22/5/2025 Continue to modify the analysis content based on the feedback from Supply Chain Executive, including optimizing the data display format, supplementing missing data fields, and reorganizing the analysis logic. While completing the modifications, also start preparing for the report to be given tomorrow and organize the required forecast model results.

 21/5/2025 Reported and presented the completed analysis tasks to Supply Chain Executive, detailing the data integration, screening logic, analysis methods, and preliminary conclusions. During the report, Executive made suggestions for improvement, such as making the coding more logical and clear. Based on these feedbacks, the report content and data processing flow were adjusted and optimized.

 20/5/2025 Continue to advance and complete the analysis tasks at hand, including collating the merged data, performing necessary screening conditions, and further extracting valuable information for sales and inventory analysis. Then make a preliminary visualization of the data for subsequent reporting. In addition, also participated in the company's evaluation and review activities for the Ecoshop product line.

 19/5/2025 I accepted a new task assigned by Supply Chain Executive and started to learn how to integrate files from three different sources and perform effective data cleaning and filtering. I also performed filtering and cleaning on the merged data, such as deleting null values, removing irrelevant SKUs, removing duplicate records, and filtering out unnecessary data under specified conditions.

 16/5/2025 In the process of using ARIMA for sales forecasting, it was found that there were significant differences in the forecast results obtained by manually setting the parameters (p, d, q) and using the parameters automatically generated by auto-ARIMA.

 15/5/2025 Successfully mastered the core concepts and application methods of the ARIMA model, including how to optimize model settings by manually adjusting parameters (p, d, q) and using automated tools (such as auto-ARIMA). Based on the optimized parameters, the ARIMA model was successfully used to predict sales data, and the accuracy and stability of the prediction were verified by comparing the actual sales volume.

 14/5/2025 After gaining a deep understanding of the working principle of the ARIMA model, I began to try to apply the ARIMA model to the prediction of sales data. By setting different parameter combinations (p, d, q), the model was trained and tested, and its prediction accuracy was evaluated. In the process of applying ARIMA, I further learned about another model that is more suitable for data with seasonal fluctuations - SARIMA (Seasonal ARIMA). SARIMA adds seasonal parameters (P, D, Q, s) to ARIMA, which can more effectively handle the impact of cyclical changes such as holidays on sales.

 13/5/2025 Continue to write sales forecast related reports, further organize and summarize the forecasting methods and model comparison results used, ensure that the content is logically clear, structurally complete, and has practical reference value. At the same time, start to systematically learn and understand the operating principle of the ARIMA (AutoRegressive Integrated Moving Average) model. Through theoretical learning combined with modeling attempts of actual sales data, you can initially understand how to choose appropriate parameters (p, d, q) and prepare for the subsequent establishment of a more accurate sales forecast model.

 12/5/2025 Public Holiday

 9/5/2025 Continue to learn Python programming. While learning the technology, I also reported to the Manager the sales forecast method I had completed before, and detailed the model type, forecasting logic, evaluation indicators (such as MAPE), and comparative analysis results. During the reporting process, I further modified and optimized the forecasting method based on the feedback and suggestions from the Manager. After completing the modification, I started to write the relevant report.

 8/5/2025 Continue to learn in depth how to write and apply various formulas in Python for data analysis, and start to pay attention to the error detection and correction process in formula calculations. In actual operations, by running different operation logics, gradually identify common problems, such as calculation errors caused by mismatched data types. Once an error is found, debug and optimize the code immediately.

 7/5/2025 Continue to learn Python data processing skills in depth, especially on the basis of mastering basic formula calculations, further learn how to add more complex formula operations and logical judgments

 6/5/2025 After mastering the basic data importing skills, learn how to use formulas and operation expressions in Python to process and analyze the imported data. Through tools such as pandas, try to use common calculation formulas such as addition, subtraction, multiplication, division, conditional judgment, average, growth rate, etc. to process various fields in the sales data, such as calculating the daily growth rate.

 5/5/2025 After analyzing and comparing various forecasting methods, it was found that the forecast results obtained using the Holt-Winters Additive model are superior to other methods in terms of accuracy (such as MAPE), especially when dealing with relatively stable seasonal fluctuations. Therefore, the Additive model is tentatively selected as the preferred solution for current sales forecasting. At the same time, in order to better automate the data analysis and forecasting process, I began to learn basic Python programming, especially basic operations related to data processing. At this stage, I have mastered how to use the polars library for file reading, such as importing sales data from local through read_csv() or read_excel() statements.

 2/5/2025 The original forecasting model based on the week was adjusted, and the Holt-Winters method and the Multiple Decomposition model were all changed to forecast sales on a "day" basis. This adjustment enables the model to more carefully capture the immediate impact of holidays, promotions and daily fluctuations on sales trends, thereby improving the sensitivity and accuracy of the forecast. After that, we had another in-depth discussion with Supply Chain Executive, and learned that the Holt-Winters method contains two different forms of seasonal processing: additive and multiplicative. Based on the understanding of the principles and applicable scenarios of both, the additive and multiplicative Holt-Winters models were used to conduct forecasting experiments on the same data, and the forecast results were compared and analyzed with the Multiple Decomposition to evaluate the performance differences of each method under different data modes.

 1/5/2025 Public Holiday