“FORECASTING OF INSULIN AND
DIABETES PREDICTION USING
MACHINE LEARNING”
-Machine Learning-based insulin forecasting
By: Hanisha SM
Ruchitha S
Geetha N
Sanjay PS
Dept. of Artificial Intelligence & Data Science
Under the Guidance of
Ms.Sushma M
Assistant professor
Department.AI & DS
INTRODUCTION
Diabetes requires continuous self-monitoring and
insulin management.
Managing food intake manually is difficult and error-
prone.
Voice assistants and AI can modernize diabetes
management.
This project offers a voice-based tool that calculates
carbs and predicts insulin dosage.
PROBLEM STATMENT
Project title:“Forecasting of Insulin and Diabetes Prediction using
machine learning”
Manual logging of food and insulin doses is inefficient.
Difficult for elderly/less tech-savvy users to operate diet based
tools.
Guesswork in insulin dosage is risky.
Existing apps & tools lack voice-enabled features and
personalized insulin prediction.
No diet/nutrition int
RESEARCH GAPS
Existing work Gaps
IoT-based monitoring No speech-based interaction
ML-based diabetes detection No dosage calculation
Voice-based triage tools No meal logging
Glucose prediction models No real-time spoken input
Voice assistants for elderly No diet/nutrition integration
LITERATURE SURVEY
UNDERSTANDING AND
AUTHORS TITILE DRAWBACKS
SOLUTIONS
Drawbacks: High
Authors: F. Enhanced detection of
computational Problem: Accessibility of technology for
Rustam Year: diabetes mellitus using
requirements for widespread use. Solution: Opti
2024 novel techniques
advanced models.
Drawbacks: Balancing Problem: Lack of trust in AI predictions
Authors: R.
Towards Transparent transparency and among healthcare providers. Solution:
Hasan Year:
Diabetes Prediction accuracy can be Use explainable AI methods to enhance
2024
challenging. trust.
Drawbacks: Genetic Problem: Models may miss critical risk
Authors: A. The Role of Genetic
data is often factors related to genetics. Solution:
Patel, N. Singh Factors in Diabetes
underutilized in Incorporate genetic information into
Year: 2024 Prediction Models
predictive models. existing models to improve accuracy.
Drawbacks: Problem: One-size-fits-all models may
Authors: A. Personalized Personalization can not be effective for individual patients.
Brown, J. Smith Diabetes Prediction complicate model Solution: Develop adaptive models that
Year: 2025 Models: A Review development and can learn from individual patient data
validation over time
OBJECTIVES
To enable spoken and typed meal input.
To analyze how insulin levels influence diabetes progression.
To use NLP to parse food items and quantities.
To estimate carb intake using nutrition databases.
To predict insulin dosage using ML models.
To store and analyze user data to track trends.
To provide a hands-free, easy-to-use experience for diabetics.
METHODOLOGY
1: Dataset Collection
Patient data (age, gender, BMI, blood glucose, insulin levels)
2: Preprocessing
Null handling, scaling, balancing classes
3: Model Building
Train ML models on features to predict insulin dosage and diabetes
probability
4: Voice Integration
Use Web Speech API → NLP (spaCy/NLP.js) to identify food +
quantity
Fetch nutrition from APIs (Nutritionix)
5: Prediction Pipeline
Calculate carbs → ML model → Predict insulin → Store in database
6: Frontend
React + Tailwind for UI, Node.js for API
DEEP LEARNING-BASED DIABETICS PREDICTION
FLOWCHART
REAL LIFE USE CASE:
A diabetic user says:
“I had 2 dosas with chutney.”
→ System converts voice to text
→ Parses food items
→ Fetches carb values
→ Predicts insulin dose
→ Displays: “Take 4 units of insulin”
→ Stores for doctor review later
EXPECTED OUTCOME
Real-time insulin dosage recommendation
Higher accuracy for insulin & diabetes
prediction
Personalized health monitoring for diabetics
Accessible voice-based solution
Trends and logs for doctor review
Scalable architecture for future enhancements
like doctor dashboard and PWA/mobile app
SUMMARY
Combines Voice Assistant, NLP, and ML in a single
tool
Bridges the gap between AI tech and real-world
diabetes care
Modern, voice-enabled, and user-friendly
Aimed to be scalable and helpful in long-term diabetes
self-management
Ideal for diabetic patients, doctors, and caregivers
Scalable, modern, and user-friendly system
Machine Learning-based insulin forecasting
REFERENCES
1. Smith, J., & Williams, E. (2020). Predicting Diabetes Risk Using Machine Learning
Algorithms. International Journal of Medical Informatics, 129, 95-103.
2. Green, S., & Lee, M. (2019). Insulin Sensitivity Prediction and Control in Type 2
Diabetes. Journal of Diabetes Science and Technology, 13(4), 867-875.
3. Karim, A., & Wong, J. (2021). Real-time Prediction of Blood Glucose Levels Using
Insulin and Lifestyle Factors. Diabetes Technology & Therapeutics, 23(1), 23-30.
4. Johnson, R., & Garcia, M. (2020). Artificial Intelligence in Insulin Dose Adjustment
for Type 1 Diabetes. Journal of Diabetes and Technology, 22(5), 721-729.
5. Brown, L., & Moore, D. (2021). Diabetes Prediction Using Wearable Sensors and
Machine Learning Models. Sensors in Healthcare, 19(3), 278-285
THANK YOU!
