🧠 Understanding Human-Emotional State

Developed a machine learning system that understands user emotions from journal text and suggests what the user should do and when.

---

⚙️ Setup Instructions

1. Install required libraries: pip install pandas numpy scikit-learn matplotlib

2. Place the files:

   • training_data.xlsx
   • testing_data.xlsx

3. Open the notebook: ML_Project.ipynb

4. Run all cells step by step

---

🧠 Approach

The system works in 3 main steps:

1. Understand emotion from text
2. Predict how strong the emotion is (intensity)
3. Suggest what action the user should take and when

---

🔧 Feature Engineering

We used two types of features:

1. Text Features

• Used TF-IDF (Term Frequency - Inverse Document Frequency)
• Converts text into numbers
• Captures important words and phrases

2. Metadata Features

• sleep_hours
• stress_level
• energy_level
• time_of_day
• ambience_type
• previous_day_mood
• face-emotion hint
• reflection quality

These are scaled using StandardScaler

---

🤖 Model Choice

We used:

Emotional State (Classification)

• Model: Linear SVM (LinearSVC)
• Reason: Works well for text classification

Intensity (Regression)

• Model: Ridge Regression
• Reason: Simple and stable for numeric prediction

---

⚡ How to Run

1. Train model using training dataset
2. Generate predictions on test dataset
3. Run decision engine
4. Create final predictions.csv

Output includes:

• predicted_state
• predicted_intensity
• confidence
• uncertain_flag
• what_to_do
• when_to_do
• supportive_message

---

📊 Output Example

State: focused
Intensity: 3
Action: deep_work
When: now

---

📌 Note

Test dataset does not have labels, so evaluation metrics like accuracy and MSE are calculated only on training data.