top of page
omarnoeman

Data Science Life Cycle (DSLC)

Updated: May 4, 2023


The software industry has seen evolution and transformation in the software Development Lifecycle. The coming days are when we talk and act more on another Lifecycle that deals with data science – Data Science Life Cycle-DSLC.

Today we run AI projects with a pre-defined set of steps to get the maximum results from AI. Here are some steps to be followed for your AI journey.



  1. Identify: Identify & define the business problem that you are trying to solve.

  2. Collect: Collecting and gathering data related to the business case.

  3. Clean: Cleaning and balance data, remove the bias.

  4. Model: Modeling using different machine learning techniques (Supervised Learning, Unsupervised Learning, Neural networks, ensemble techniques, deep learning, NLP etc.)

  5. Evaluate: Evaluation of AI model to reach the optimum model performance.

  6. Deploy: Deployment of a model in production.

  7. Optimize: Optimization and upgrading AI model.

As Organizations get more matured in the AI journey, it is very important to know and understand the various challenges one may experience. AI journey will never be a smooth journey, and it is all about Predictions, Corrections & Improvements. During this bumpy ride, one should be prepared to face the challenges across the journey. Some of the main challenges are

  1. Data privacy: data might be available, but the organization’s regulations comply with preventing them from using this data for Machine learning, especially if they plan to use Cloud services for visualization and modelling.

  2. Bias in data: in most cases, you will have unbalanced data that are waited toward a certain type or certain category over other categories, requiring more data balancing and cleaning efforts.

  3. Transparency: AI will form future humanity and require an ethical obligation to be transparent in all efforts from organization and developers.

  4. Scalability: Ai solution can be made available and do certain job level but most probably will fail after a certain level of deployment, which requires an organization to build a scalable solution for enterprise applications.

  5. Model drift: degradation of a model’s prediction power due to changes in the environment will create problems in operation, which require monitoring to model accuracy

  6. Security: One of the biggest hurdles in securing machine learning systems is that data in machine learning systems play an outside role in security.

9 views0 comments

Comments


bottom of page