Page 222 - Ai_417_V3.0_C9_Flipbook
P. 222

✶ Preprocess  data: Simplify numerical features, convert categorical variables, and create new metrics like
                   customer tenure.
                 • Modelling
                    ✶ Select techniques: Choose machine learning algorithms suitable for classification tasks, such as logistic
                   regression, decision trees, and random forests.

                    ✶ Train models: Use the prepared data to train multiple models, adjusting hyperparameters and performing
                   cross-validation to optimise performance.
                 • Evaluation

                    ✶ Evaluate models: Assess the performance of each model using metrics like accuracy, precision, recall, and
                   F1-score.
                    ✶ Compare models: Compare the performance of different models to select the best-performing one for
                   deployment.
                 • Deployment

                    ✶ Deploy model: Integrate the selected model into the company's customer management system to predict
                   churn risk for new customers.
                    ✶ Monitor performance: Monitor the model's predictions in real-time, track churn rates, and gather feedback
                   from customer service interactions.

              In this example, each phase of the AI project cycle builds upon the outputs of the previous phase:

                                                 AI Project Cycle Mapping Template

                                         Data            Data
                Problem Scoping                                        Modelling        Evaluation      Deployment
                                     Acquisition      Exploration
               The                 Gather customer  Analyse          Select machine   Evaluate        Integrate the
               telecommunications  demographics,    customer         learning        each model's     model to predict
               company wants to    usage patterns,   demographics,   algorithms for   performance     new customer
               reduce customer     service history,   usage patterns,   classification,   using accuracy,   churn risk.
               churn rates.        and churn        and churn        like logistic   precision, recall,
                                   status data      rates with       regression,     and F1-score.
                                   from company     visualisations   decision trees,
                                   databases.       and statistical   and random
                                                    summaries.       forests.


                 • Problem Scoping: States the problem that needs attention.
                 • Data Acquisition: Data acquisition consists of two words: Data and Acquisition. Data refers to the raw facts,
                figures, information, or statistics; where as, acquisition refers to acquiring data for the project.
                 • Data Exploration: It is the first step of data analysis that is used to visualise data. It generates insights that are
                used to inform modelling decisions.
                 • Modelling: Develops predictive models based on insights gained from data exploration.
                 • Evaluation: Assesses model performance by feeding the data into the model and comparing the output with
                the actual answers. It is used to determine the best model for deployment.
                 • Deployment: Integrates the selected model into the company's systems for real-world usage.



                    220     Touchpad Artificial Intelligence (Ver. 3.0)-IX
   217   218   219   220   221   222   223   224   225   226   227