The second-largest source of income in Sri Lanka is the apparel industry. While lots of companies in this industry are gradually adopting new technologies like smart quality management systems, even some of the largest companies in Sri Lanka are still taking a very traditional approach in their product management. Even the slightest improvement in the efficiency and reliability of this process with seamless integration of new technologies like Big Data and ML could go a long way in strengthening the Sri Lankan economy. There is no better place than Hirdaramani Apparel, the third largest and the oldest apparel exporter in Sri Lanka, to take a big step in this direction. The possibility to make a contribution to such a shift is thrilling, especially with a project that has virtually no cost, yet promising results.
There are KPIs in the apparel industry such as Cut-to-ship ratio, On-Time Delivery rate, Man to machine ratio, average style changes over time, etc. The problem we are trying to solve is the issues in achieving on-time delivery.
Unreliable prediction of delivery Unpredictable delivery of raw materials Traditional/manual analysis and planning - (e.g.: excel sheets) No user-friendly system to visualize and understand the data Data from past orders are not properly used for the future (short term analysis)
To develop an application that allows the user to load datasets and see meaningful information and visualizations in an easy-to-understand manner. Administrators will be able to pay attention to the underlying causes of low OTD by identifying them with the application. Furthermore, it can predict the OTD based on the fields involved.
- An application to input data sets and get outputs/visualizations
- To determine the causes of low OTD
- To predict the OTD based on the fields involved
- The level of customer satisfaction
- Improve the stability of the process
- Early detection of problems
- Data analysis - ML-based analysis is faster, more accurate, reliable than traditional approaches.
- The accuracy of the prediction models is measurable. The expected users of the application are the Administration and the Management staff, hence the solution should consist of an easy-to-use user interface with easily interpretable outputs/visualizations of data analysis and prediction.
- Find the factors that affect OTD, OTD related performance measures.
- Proof of Concept
- Create a software tool to insert and present information
- Extensibility to incorporate more data & analysis in the future
- Developing Sri Lanka's apparel industry to its full potential, building on its strengths, such as its long-time expertise and skillset
- Gathering data and analyzing the problem.
- Evaluate which aspects of the data contribute most to delivery time.
- Utilize open-source tools and state-of-the-art machine learning techniques to analyze data related to OTD.
- Trying different ML modeling techniques to identify the most optimal model.
- Based on the results of the data analysis, determine what needs to be included in the software application.
- Present the findings in an intuitive user interface.
- Design and develop the software components.
- Integration of the components.
- Testing
- Review the results and revise the plan if necessary.
The use of well-known platforms makes it possible to take advantage of algorithms that have already been implemented and optimized for use in dynamic applications.
- A month-by-month breakdown of the data is currently available. However, the data corresponding to individual orders would be more reliable for predicting OTD.
- Instead of a performance summary, more raw data with timestamps would be more helpful.
- Set milestones ourselves
- Statistically analyze/compare
- Measure performance for the test set