Personalized activity and diet recommendation system using machine learning

Abstract

In Sri Lankan, working professionals are generally predisposed to non-communicable illnesses because of their hectic schedules and disorderly eating habits. Less physical activity is a result of sedentary work habits, which have an impact on general health and wellbeing. By creating an intelligent system that suggests simple physical exercises and an appropriate diet that can be followed from home, the primary goal of this project is to discover answers to these problems. There are many applications that are available as to recommend a diet and exercise schedules but they are seen separately as diet plan projects and exercise schedules projects. This project will be created for active working professionals living in western provinces. Although there are many such projects, this is the project that recommends both a diet and an exercise schedule based only on the busy people of the Western Province. After applying three distinct classifiers (k-Nearest Neighbors, Random Forest, and Decision Tree) to your data, the preliminary findings. Here is a quick quantitative analysis of the classification problem, complete with confusion measures and AUC-ROC scores. Although the KNN classifier had reasonable accuracy, recall was problematic, especially when it came to categorizing one of the classes as the low recall score suggests. The precision shows that the model is mostly accurate when it predicts a good outcome. The low recall suggests that the model needs to be improved to accurately identify all positive events. In comparison to KNN, the Random Forest classifier fared better, achieving higher F1 score, accuracy, precision, and recall. Diet has 75% model accuracy and exercise has 90% model accuracy. For class 2, it received a high recall score, demonstrating improved ability to recognize favorable examples. Additionally, the model displays balanced recall and precision ratings for every class. With the lowest accuracy, precision, recall, and F1 score among the three models, the Decision Tree classifier performed the worst. To increase this classifier's performance, more fine-tuning or feature engineering may be required.