I have developed a Depression Detection App that analyzes mental health conditions using machine learning. This app is trained on a dataset containing over 23,000 data points, focusing on various factors such as sleep patterns, appetite, social interactions, academic pressure, and self-perception. The model categorizes depression levels into four categories: Low, Mild, Moderate, and High. It specifically targets individuals aged 13-25, which includes teenagers and young adults, as this group is particularly vulnerable to mental health challenges. By using advanced machine learning algorithms, the app helps in detecting early signs of depression and provides users with insightful results based on their responses.
To build the model, I have utilized several machine learning techniques, including:
- Data Preprocessing: StandardScaler and LabelEncoder are used to standardize and encode categorical data.
- Data Splitting: The dataset is split into training and testing sets to ensure the model generalizes well.
- SMOTE (Synthetic Minority Over-sampling Technique): Applied to balance the dataset and avoid bias toward majority classes.
- Machine Learning Models: Various classifiers, including Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest, Decision Tree, Gaussian Naive Bayes, and XGBoost, are trained to predict depression levels accurately.
- Performance Metrics: The models are evaluated using accuracy, F1-score, precision, recall, and confusion matrix analysis to determine their effectiveness.
- Data Visualization: Graphs and bar charts are used to illustrate the performance of different models, making it easier to compare results.
Among the models tested, the Support Vector Machine (SVM) demonstrated the highest accuracy and overall performance, making it the best choice for depression detection. The XGBoost model also performed well, showcasing strong predictive capabilities. On the other hand, the Decision Tree and Random Forest models had comparatively lower accuracy, indicating their limitations in handling complex patterns in the data. The Gaussian Naive Bayes model provided a balanced trade-off between accuracy and computational efficiency. Overall, the model selection was driven by a combination of performance metrics, ensuring the most reliable predictions for users.
Currently, I am working on implementing a feedback filtering model that will categorize user reviews into positive and negative feedback. The system will function as follows:
- Good Reviews: Automatically added to the website to showcase positive feedback.
- Negative Reviews: First reviewed by me before being published. I will have the opportunity to respond to these reviews before making them publicly visible.
This feature will help in improving user engagement and addressing concerns effectively. By leveraging Natural Language Processing (NLP) techniques, the system will intelligently analyze user feedback and take appropriate actions accordingly.
The depression detection model is deployed using Flask, a lightweight web framework. The web interface allows users to input their details and receive an analysis of their mental health condition. The application consists of:
- Homepage: Provides information about the app.
- Depression Test Page: Users can enter their details, and the model predicts their depression level.
- Results Page: Displays the predicted depression category.
The model is saved and loaded using Joblib, ensuring efficient deployment and usability.
Going forward, I plan to:
- Improve the accuracy of the depression detection model by incorporating more sophisticated deep learning techniques.
- Enhance the feedback system by integrating sentiment analysis and chatbot functionality for automated responses.
- Expand the dataset with real-world data to make predictions more reliable and robust.
This project aims to make mental health assessments more accessible and improve online user interactions through intelligent feedback filtering. Let me know if you have any suggestions or improvements!