At-Risk Prediction Model for School Campus

Aug 12

Student Achievement

Outline

Predict State Accountability Rating

Problem: School is rated “Low-Performing” for five consecutive years
Solution: Create an at-risk prediction model
Tools: Microsoft Excel, Tableau, JMP Pro 16

Client

A school campus that needs to increase student achievement is seeking a prediction model to identify at-risk students and calculate accountability ratings. They need a system of identification that is robust and accurate. This information will guide intervention plans targeting low performing students. This educational institution collaborated with Tahn Analytics to design a prediction model.

Problem: School is rated “Low-Performing” for five consecutive years

Paired sample t-Test, classification and prediction models are applied to forecast pass/fail outcomes and estimate summative assessment percentage scores. Evaluations are performed to rank and determine best fit models. Exploratory data analysis is performed to acquire historical knowledge of campus assessment data, identify trends, patterns, and outliers.

Process

Retrieve campus assessment data from district database
Data Cleaning
Data Visualization
Training and evaluation of Machine Learning Models

Solution: Create an At-Risk Prediction Model

To identify at-risk students, we utilize ML models to classify and predict:

Paired Sample T-Test to establish that interim assessments scores are not equivalent to summative assessment scores
Logistic Regression, Classification Trees, Neural Networks – Models to predict pass/fail outcomes
Linear Regression, Decision Trees, Neural Networks – Models to predict summative assessment scores

Result: Pass/Fail Prediction Model for Reading Summative Assessments

Originally the client believed interim assessments were time consuming and futile. This analysis informs leaders on the validity of at-risk indicators supplied by interim assessments. In conclusion the prediction model was able to identify at-risk students with 91% accuracy. Thinking forward we suggest the campus collect data on methods of intervention. These variables include effective teacher modeling, guided practice, checks for understanding, feedback to learners, independent practice, varied instructional activities, student engagement, students verbalizing their learning, physical and visual representations or models of number concepts or problem-solving situations. Inclusion of these variables will improve the accuracy, velocity, and impact of the prediction model.

Mujidat Saaka