app.py

import streamlit as st
import pandas as pd
import plotly.express as px
import plotly.graph_objects as go
from datetime import datetime, timedelta

# =================== CONFIG =====================
st.set_page_config(
    page_title="MineVision AI - Advanced Fatigue Analytics",
    page_icon="⛏️",
    layout="wide",
    initial_sidebar_state="expanded"
)

# Custom CSS for professional look
st.markdown("""
<style>
    .main-header {
        background-color: #003366;
        padding: 20px;
        border-radius: 10px;
        color: white;
        text-align: center;
        margin-bottom: 20px;
    }
    .metric-card {
        background-color: #f0f8ff;
        padding: 15px;
        border-radius: 8px;
        border-left: 5px solid #003366;
    }
    .insight-box {
        background-color: #f9f9f9;
        padding: 15px;
        border-radius: 8px;
        border-left: 5px solid #ff6b6b;
        margin: 10px 0;
    }
    .footer {
        text-align: center;
        padding: 20px;
        color: gray;
        font-size: 0.9em;
    }
    .risk-matrix {
        border-collapse: collapse;
        width: 100%;
        margin: 20px 0;
    }
    .risk-matrix th, .risk-matrix td {
        border: 1px solid #ddd;
        padding: 8px;
        text-align: left;
    }
    .risk-matrix th {
        background-color: #f2f2f2;
    }
    .critical { background-color: #ffcccc; }
    .high { background-color: #ffebcc; }
    .medium { background-color: #ffffcc; }
    .low { background-color: #e6ffe6; }
    .chat-container {
        background-color: white;
        padding: 15px;
        border-radius: 10px;
        height: 400px;
        overflow-y: auto;
        margin-top: 20px;
        border: 1px solid #ccc;
    }
    .user-message {
        background-color: #e3f2fd;
        color: black;
        padding: 10px;
        border-radius: 10px;
        margin: 10px 0;
        text-align: right;
        border: 1px solid #bbdefb;
    }
    .ai-message {
        background-color: #f5f5f5;
        color: black;
        padding: 10px;
        border-radius: 10px;
        margin: 10px 0;
        text-align: left;
        border: 1px solid #e0e0e0;
    }
    .chat-box {
        background-color: white;
        border: 1px solid #ccc;
        border-radius: 8px;
        padding: 10px;
        margin-top: 10px;
        width: 100%;
    }
    .user-question {
        background-color: white;
        border: 1px solid #ccc;
        border-radius: 8px;
        padding: 10px;
        margin-bottom: 10px;
    }
    .ai-answer {
        background-color: #f0f0f0;
        border: 1px solid #ccc;
        border-radius: 8px;
        padding: 10px;
        margin-bottom: 10px;
    }
</style>
""", unsafe_allow_html=True)

# Header
st.markdown('<div class="main-header"><h1>Safety Analysis and AI - Advanced Fatigue Analysis</h1><p>Proactive Safety Intelligence for Mining Operations</p></div>', unsafe_allow_html=True)

# =================== CHAT AI SECTION =====================
st.subheader("MineVision AI Assistant")

# Initialize session state for chat
if 'chat_history' not in st.session_state:
    st.session_state.chat_history = []

# Display chat history in a fancy box with white background
st.markdown('<div class="chat-container">', unsafe_allow_html=True)
for message in st.session_state.chat_history:
    if message['role'] == 'user':
        st.markdown(f'<div class="user-message">You: {message["content"]}</div>', unsafe_allow_html=True)
    else:
        st.markdown(f'<div class="ai-message">MineVision AI: {message["content"]}</div>', unsafe_allow_html=True)
st.markdown('</div>', unsafe_allow_html=True)

# Input for user question
user_input = st.text_input("Ask a question about the fatigue data...", key="chat_input")

if st.button("Send", key="send_button"):
    if user_input:
        # Add user message to history
        st.session_state.chat_history.append({"role": "user", "content": user_input})
        
        # Process the question and generate response based on data
        response = ""
        user_input_lower = user_input.lower()
        
        # Improved RAG responses based on data analysis and Wenco insights
        if "operator" in user_input_lower and ("sering" in user_input_lower or "banyak" in user_input_lower or "most" in user_input_lower or "highest" in user_input_lower):
            if col_operator and not df.empty:
                top_operator = df[col_operator].value_counts().idxmax()
                count = df[col_operator].value_counts().iloc[0]
                total_alerts = len(df)
                percentage = (count / total_alerts) * 100
                response = f"Operator dengan jumlah kejadian ngantuk paling banyak adalah **{top_operator}** dengan **{count}** kejadian ({percentage:.1f}% dari total {total_alerts} kejadian)."
            else:
                response = "Tidak ada data operator yang tersedia."
        elif "shift" in user_input_lower and ("banyak" in user_input_lower or "most" in user_input_lower or "highest" in user_input_lower):
            if col_shift and not df.empty:
                top_shift = df[col_shift].value_counts().idxmax()
                count = df[col_shift].value_counts().iloc[0]
                total_alerts = len(df)
                percentage = (count / total_alerts) * 100
                response = f"Shift dengan jumlah kejadian ngantuk paling banyak adalah **Shift {top_shift}** dengan **{count}** kejadian ({percentage:.1f}% dari total {total_alerts} kejadian)."
            else:
                response = "Tidak ada data shift yang tersedia."
        elif "jam" in user_input_lower and ("banyak" in user_input_lower or "most" in user_input_lower or "highest" in user_input_lower or "sering" in user_input_lower):
            if "hour" in df.columns and not df.empty:
                top_hour = df["hour"].value_counts().idxmax()
                count = df["hour"].value_counts().iloc[0]
                total_alerts = len(df)
                percentage = (count / total_alerts) * 100
                response = f"Jam dengan jumlah kejadian ngantuk paling banyak adalah pukul **{top_hour}:00** dengan **{count}** kejadian ({percentage:.1f}% dari total {total_alerts} kejadian)."
            else:
                response = "Tidak ada data jam yang tersedia."
        elif "fleet" in user_input_lower and ("banyak" in user_input_lower or "most" in user_input_lower or "highest" in user_input_lower):
            if col_fleet_type and not df.empty:
                top_fleet = df[col_fleet_type].value_counts().idxmax()
                count = df[col_fleet_type].value_counts().iloc[0]
                total_alerts = len(df)
                percentage = (count / total_alerts) * 100
                response = f"Fleet type dengan jumlah kejadian ngantuk paling banyak adalah **{top_fleet}** dengan **{count}** kejadian ({percentage:.1f}% dari total {total_alerts} kejadian)."
            else:
                response = "Tidak ada data fleet type yang tersedia."
        elif "total" in user_input_lower and "alert" in user_input_lower:
            response = f"Total kejadian fatigue alert adalah **{len(df)}**."
        elif "average" in user_input_lower and ("duration" in user_input_lower or "lama" in user_input_lower):
            if "duration_sec" in df.columns and not df.empty:
                avg_duration = df["duration_sec"].mean()
                response = f"Rata-rata durasi kejadian fatigue adalah **{avg_duration:.2f} detik**."
            else:
                response = "Tidak ada data durasi yang tersedia."
        elif "risk" in user_input_lower and ("category" in user_input_lower or "level" in user_input_lower):
            if 'risk_category' in df.columns and not df.empty:
                risk_counts = df['risk_category'].value_counts()
                total_alerts = len(df)
                response = f"Kategori risiko kelelahan:\n"
                for category, count in risk_counts.items():
                    percentage = (count / total_alerts) * 100
                    response += f"- {category}: {count} kejadian ({percentage:.1f}% dari total)\n"
            else:
                response = "Tidak ada data kategori risiko yang tersedia."
        elif "speed" in user_input_lower and ("high" in user_input_lower or "fast" in user_input_lower):
            if col_speed and not df.empty:
                high_speed_threshold = df[col_speed].quantile(0.75)
                high_speed_count = len(df[df[col_speed] >= high_speed_threshold])
                total_alerts = len(df)
                percentage = (high_speed_count / total_alerts) * 100
                response = f"Jumlah kejadian fatigue pada kecepatan tinggi (> {high_speed_threshold:.0f} km/h) adalah **{high_speed_count}** kejadian ({percentage:.1f}% dari total {total_alerts} kejadian)."
            else:
                response = "Tidak ada data kecepatan yang tersedia."
        elif "critical" in user_input_lower and "hour" in user_input_lower:
            critical_hours = [2, 3, 4, 5]
            critical_alerts = df[df['hour'].isin(critical_hours)]
            total_alerts = len(df)
            percentage = (len(critical_alerts) / total_alerts) * 100 if total_alerts > 0 else 0
            response = f"Jumlah kejadian fatigue pada jam kritis (2-5 AM) adalah **{len(critical_alerts)}** kejadian ({percentage:.1f}% dari total {total_alerts} kejadian)."
        elif "madar" in user_input_lower:
            if col_operator and not df.empty:
                # Check if "Madar" is an operator in the data
                madar_data = df[df[col_operator].str.contains('Madar', case=False, na=False)]
                if not madar_data.empty:
                    madar_count = len(madar_data)
                    total_alerts = len(df)
                    percentage = (madar_count / total_alerts) * 100
                    response = f"Operator **Madar** tercatat memiliki **{madar_count}** kejadian ngantuk ({percentage:.1f}% dari total {total_alerts} kejadian)."
                else:
                    response = "Operator 'Madar' tidak ditemukan dalam data."
            else:
                response = "Tidak ada data operator yang tersedia."
        elif "frms" in user_input_lower or "fatigue risk management" in user_input_lower:
            response = "Sistem FRMS (Fatigue Risk Management System) yang digunakan adalah pendekatan berbasis data untuk mengidentifikasi, menilai, dan mengendalikan risiko kelelahan. Sistem ini menggabungkan data dari berbagai sumber seperti jam kerja, pola tidur, dan deteksi kelelahan langsung untuk memberikan wawasan dan rekomendasi pencegahan. Dalam konteks mining, FRMS membantu mengurangi kecelakaan dan meningkatkan produktivitas dengan mengelola faktor-faktor risiko kelelahan secara sistematis."
        elif "intervensi" in user_input_lower or "intervention" in user_input_lower:
            # Use data to explain intervention rates
            if col_operator and not df.empty:
                total_operators = df[col_operator].nunique()
                total_alerts = len(df)
                # Assuming each alert might require an intervention
                avg_interventions_per_operator = total_alerts / total_operators if total_operators > 0 else 0
                response = f"Berdasarkan data, rata-rata intervensi yang diperlukan per operator adalah sekitar **{avg_interventions_per_operator:.2f}** kejadian. Menurut dokumentasi Wenco, rata-rata hanya ada satu alarm per 22 jam operator, yang menunjukkan tingkat intervensi yang dapat dikelola."
            else:
                response = "Data untuk menghitung tingkat intervensi tidak tersedia."
        elif "implementasi" in user_input_lower or "implementation" in user_input_lower or "resistensi" in user_input_lower or "resistance" in user_input_lower:
            response = "Berdasarkan dokumentasi Wenco, implementasi FRMS di industri mining menghadapi beberapa tantangan seperti resistensi tenaga kerja (privasi, takut dikenai sanksi), isu teknis (reliabilitas awal, lingkungan keras), dan hambatan manajemen (biaya tinggi, justifikasi ROI). Namun, realitas modern menunjukkan bahwa kekhawatiran seperti 'fleet shutdown' berlebihan, dengan tingkat alarm yang dapat dikelola. Vendor juga telah berkembang, menawarkan dukungan dan model penerapan yang lebih baik."
        else:
            # Improved fallback response with more context
            context_info = []
            if col_operator:
                context_info.append(f"Operator: {df[col_operator].nunique() if not df.empty else 0} unik")
            if col_shift:
                context_info.append(f"Shift: {sorted(df[col_shift].dropna().unique()) if not df.empty else []}")
            if "hour" in df.columns:
                context_info.append(f"Jam: {min(df['hour']) if not df.empty and not df['hour'].isna().all() else 0}-{max(df['hour']) if not df.empty and not df['hour'].isna().all() else 23}")
            if col_fleet_type:
                context_info.append(f"Fleet: {df[col_fleet_type].nunique() if not df.empty else 0} jenis")
            if "duration_sec" in df.columns:
                context_info.append(f"Durasi: rata-rata {df['duration_sec'].mean():.2f} detik")
            if col_speed:
                context_info.append(f"Kecepatan: hingga {df[col_speed].max() if not df.empty and not df[col_speed].isna().all() else 0} km/h")
            
            context_str = ", ".join(context_info)
            response = f"Pertanyaan Anda tidak dapat diproses. Silakan tanyakan tentang operator, shift, jam, fleet type, total alert, durasi, kategori risiko, kecepatan tinggi, jam kritis, FRMS, intervensi, atau implementasi. Data saat ini mencakup: {context_str}."
        
        # Add AI response to history
        st.session_state.chat_history.append({"role": "assistant", "content": response})
        
        # Rerun to update the chat display
        st.rerun()


# =================== LOAD DATA ======================
@st.cache_data
def load_data():
    # Load data from the uploaded file
    try:
        df = pd.read_excel('manual fatique.xlsx', sheet_name=None, engine="openpyxl")
        
        # If the file has multiple sheets, concatenate them
        if isinstance(df, dict):
            df = pd.concat(df.values(), ignore_index=True)
        
        df.columns = df.columns.astype(str).str.strip().str.lower().str.replace(" ", "_")

        # auto detect important columns
        col_operator = next((c for c in df.columns if "operator" in c or "driver" in c), None)
        col_shift = next((c for c in df.columns if "shift" in c), None)
        col_asset = next((c for c in df.columns if "asset" in c or "vehicle" in c or "fleet" in c), None)
        col_fleet_type = next((c for c in df.columns if "parent_fleet" in c), None)
        col_speed = next((c for c in df.columns if "speed" in c or "km/h" in c), None)

        # detect timestamps (using the actual column names from the provided file)
        start_time_cols = [c for c in df.columns if "gmt" in c.lower() and "wita" in c.lower()]
        # Assuming the first one is start and the second is end
        if len(start_time_cols) >= 2:
            df["start"] = pd.to_datetime(df[start_time_cols[0]], errors="coerce")
            df["end"] = pd.to_datetime(df[start_time_cols[1]], errors="coerce")
        elif len(start_time_cols) == 1:
            # If only one time column, assume it's start time and set end time to start + 1 minute as a placeholder
            df["start"] = pd.to_datetime(df[start_time_cols[0]], errors="coerce")
            df["end"] = df["start"] + pd.Timedelta(minutes=1)

        df["duration_sec"] = (df["end"] - df["start"]).dt.total_seconds()
        df["hour"] = df["start"].dt.hour
        df["date"] = df["start"].dt.date # Add date column for filtering
        df["day_of_week"] = df["start"].dt.day_name() # Add day of week for analysis
        df["week"] = df["start"].dt.isocalendar().week # Add week for trend analysis
        df["month"] = df["start"].dt.month # Add month for filtering
        df["year"] = df["start"].dt.year # Add year for filtering

        # Ensure shift is integer type and handle potential decimal values by rounding
        if col_shift:
            # Convert to numeric, then round to nearest integer, then convert to int64 to remove decimals
            df[col_shift] = pd.to_numeric(df[col_shift], errors='coerce').round().astype('Int64')

        return df, col_operator, col_shift, col_asset, col_fleet_type, col_speed
    except FileNotFoundError:
        st.error("File 'manual fatique.xlsx' not found. Please check the file path.")
        return pd.DataFrame(), None, None, None, None, None
    except Exception as e:
        st.error(f"Error loading  {e}")
        return pd.DataFrame(), None, None, None, None, None


df, col_operator, col_shift, col_asset, col_fleet_type, col_speed = load_data()

if df.empty:
    st.stop()

st.success("Data Loaded Successfully")

# =================== FILTERS (Sidebar) =====================
st.sidebar.header("Filters")

# Year Filter
if 'year' in df.columns:
    all_years = sorted(df['year'].dropna().unique())
    selected_years = st.sidebar.multiselect(
        "Select Year (Leave blank for All)",
        options=all_years,
        default=all_years  # Default to all if none selected
    )
    if selected_years:
        df = df[df['year'].isin(selected_years)]

# Month Filter
if 'month' in df.columns:
    all_months = sorted(df['month'].dropna().unique())
    selected_months = st.sidebar.multiselect(
        "Select Month (Leave blank for All)",
        options=all_months,
        default=all_months  # Default to all if none selected
    )
    if selected_months:
        df = df[df['month'].isin(selected_months)]

# Week Filter
if 'week' in df.columns:
    all_weeks = sorted(df['week'].dropna().unique())
    selected_weeks = st.sidebar.multiselect(
        "Select Week (Leave blank for All)",
        options=all_weeks,
        default=all_weeks  # Default to all if none selected
    )
    if selected_weeks:
        df = df[df['week'].isin(selected_weeks)]

# Date Range Filter: Default to "All" if no specific range is selected
if 'date' in df.columns:
    min_date = df['date'].min()
    max_date = df['date'].max()
    # Set default value to the full range initially
    date_range_default = (min_date, max_date)

    date_range_input = st.sidebar.date_input(
        "Select Date Range (Leave blank for All)",
        value=date_range_default, # Default to full range
        min_value=min_date,
        max_value=max_date
    )

    # Check if date_range_input is empty (user cleared the dates) or default full range is kept without interaction
    if not date_range_input or (len(date_range_input) == 2 and date_range_input[0] == min_date and date_range_input[1] == max_date):
        # If empty tuple or default full range, set to actual full range and mark as not explicitly filtered
        date_range = (min_date, max_date)
        date_filtered = False
    else:
        # If user selected a specific range, use it
        date_range = tuple(date_range_input)
        date_filtered = True
    # Apply date filter
    df = df[(df['date'] >= date_range[0]) & (df['date'] <= date_range[1])]

# Operator Filter (with search functionality)
if col_operator:
    all_operators = sorted(df[col_operator].dropna().unique())
    # Use multiselect with search functionality
    selected_operators = st.sidebar.multiselect(
        f"Select {col_operator.replace('_', ' ').title()} (Leave blank for All)",
        options=all_operators,
        default=all_operators,  # Default to all if none selected
        format_func=lambda x: x  # Format function for better display
    )
    if selected_operators:
        df = df[df[col_operator].isin(selected_operators)]

# Shift Filter (with search functionality) - Ensure integers
if col_shift:
    all_shifts = sorted(df[col_shift].dropna().unique())
    # Use multiselect with search functionality
    selected_shifts = st.sidebar.multiselect(
        f"Select {col_shift.replace('_', ' ').title()} (Leave blank for All)",
        options=all_shifts,
        default=all_shifts,  # Default to all if none selected
    )
    if selected_shifts:
        df = df[df[col_shift].isin(selected_shifts)]

# Hour Range Filter
all_hours = sorted(df['hour'].dropna().unique())
if len(all_hours) > 0:
    hour_range = st.sidebar.slider(
        "Select Hour Range (Leave at full range for All)",
        min_value=int(min(all_hours)),
        max_value=int(max(all_hours)),
        value=(int(min(all_hours)), int(max(all_hours))),
        step=1
    )
    if hour_range != (int(min(all_hours)), int(max(all_hours))):
        df = df[(df['hour'] >= hour_range[0]) & (df['hour'] <= hour_range[1])]
else:
    # Handle case where there are no hours
    st.sidebar.text("No hour data available")
    hour_range = (0, 23)


# =================== FATIGUE RISK CATEGORIZATION =====================
st.subheader("Fatigue Risk Categorization")

# Define risk categories based on the provided matrix
if col_speed and "hour" in df.columns:
    # Create risk category column based on the matrix
    df['risk_category'] = df.apply(lambda row: 
        'Critical' if (row[col_speed] > df[col_speed].quantile(0.75) and row['hour'] in [2, 3, 4, 5]) else
        'High' if (row[col_speed] > df[col_speed].quantile(0.5) and row['hour'] in [2, 3, 4, 5]) else
        'Medium' if (row[col_speed] > df[col_speed].quantile(0.25) and row['hour'] in [2, 3, 4, 5]) else
        'Low' if (row[col_speed] <= df[col_speed].quantile(0.25) and row['hour'] not in [2, 3, 4, 5]) else
        'Medium', axis=1)  # Default to medium for other cases
    
    # Count alerts by risk category
    risk_counts = df['risk_category'].value_counts().reindex(['Critical', 'High', 'Medium', 'Low'])
    
    # Create a bar chart showing the distribution of risk categories
    fig_risk = px.bar(
        x=risk_counts.index,
        y=risk_counts.values,
        title="Fatigue Risk Categories Distribution",
        labels={'x': 'Risk Category', 'y': 'Number of Alerts'},
        color=risk_counts.index,
        color_discrete_map={'Critical': 'red', 'High': 'orange', 'Medium': 'yellow', 'Low': 'green'}
    )
    fig_risk.update_layout(
        xaxis_title="Risk Category",
        yaxis_title="Number of Alerts",
        height=400
    )
    # Add legend to explain each category
    fig_risk.update_layout(
        legend_title_text="Risk Level",
        legend=dict(
            orientation="v",
            yanchor="top",
            y=1,
            xanchor="left",
            x=1.02
        )
    )
    # Add annotations to explain what each risk level means
    for i, (cat, count) in enumerate(risk_counts.items()):
        if cat == 'Critical':
            fig_risk.add_annotation(
                x=cat,
                y=count + 1,
                text="High fatigue + high-speed haul road",
                showarrow=False,
                font=dict(size=10),
                bgcolor="red",
                opacity=0.8
            )
        elif cat == 'High':
            fig_risk.add_annotation(
                x=cat,
                y=count + 1,
                text="Moderate fatigue + decline haul road",
                showarrow=False,
                font=dict(size=10),
                bgcolor="orange",
                opacity=0.8
            )
        elif cat == 'Medium':
            fig_risk.add_annotation(
                x=cat,
                y=count + 1,
                text="High fatigue + low-risk task",
                showarrow=False,
                font=dict(size=10),
                bgcolor="yellow",
                opacity=0.8
            )
        elif cat == 'Low':
            fig_risk.add_annotation(
                x=cat,
                y=count + 1,
                text="Low fatigue + non-hazard task",
                showarrow=False,
                font=dict(size=10),
                bgcolor="green",
                opacity=0.8
            )
    
    st.plotly_chart(fig_risk, width="stretch")


# =================== KPI METRICS =====================
st.subheader("Executive Safety Dashboard")

col1, col2, col3, col4 = st.columns(4)

col1.metric("Total Alerts", f"{len(df):,}")
col2.metric("Operators", df[col_operator].nunique() if col_operator else "-")
col3.metric("Qty Equipment", df[col_asset].nunique() if col_asset else "-")  # Changed from "Assets" to "Qty Equipment"
col4.metric("Avg Duration (sec)", round(df["duration_sec"].mean(),2) if "duration_sec" in df.columns else "N/A")


# =================== TREND ANALYTICS =====================
st.subheader("Fatigue Trend Analysis")

# Hourly
fig_hour = px.bar(
    df.groupby("hour").size().reset_index(name="alerts"),
    x="hour", y="alerts",
    title="Fatigue Alerts by Hour"
)
st.plotly_chart(fig_hour, width="stretch")

# Shift-Based
if col_shift:
    fig_shift = px.bar(
        df.groupby(col_shift).size().reset_index(name="alerts"),
        x=col_shift, y="alerts",
        title="Fatigue Distribution by Shift"
    )
    # Force the x-axis (shift) to be categorical to avoid decimal labels
    fig_shift.update_xaxes(type='category')
    st.plotly_chart(fig_shift, width="stretch")

    # hour inside shift heatmap
    heat_df = df.groupby([col_shift, "hour"]).size().reset_index(name="alerts")

    fig_heat = px.density_heatmap(
        heat_df,
        x="hour", y=col_shift, z="alerts",
        title="Heatmap Fatigue by Shift & Hour",
        color_continuous_scale="reds"
    )
    # Force the y-axis (shift) to be categorical to avoid decimal labels
    fig_heat.update_yaxes(type='category')
    st.plotly_chart(fig_heat, width="stretch")


# Operator Ranking
if col_operator:
    operator_counts = df[col_operator].value_counts().reset_index()
    operator_counts.columns = ["operator", "alerts"]
    fig_operator = px.bar(
        operator_counts,
        x="operator", y="alerts",
        title="Top Fatigue Alerts by Operator"
    )
    st.plotly_chart(fig_operator, width="stretch")


# =================== NEW CHARTS (Based on Mining Fatigue Factors) =====================
st.subheader("Advanced Mining Fatigue Analytics")

# 1. Day of Week Analysis (Workload Pattern)
if 'day_of_week' in df.columns:
    day_counts = df['day_of_week'].value_counts().reindex(['Monday', 'Tuesday', 'Wednesday', 'Thursday', 'Friday', 'Saturday', 'Sunday'])
    fig_day = px.bar(
        day_counts,
        x=day_counts.index, y=day_counts.values,
        title="Fatigue Alerts by Day of Week (Workload Pattern)"
    )
    st.plotly_chart(fig_day, width="stretch")

# 2. Fleet Type Analysis (Task & Workload)
if col_fleet_type:
    fleet_counts = df[col_fleet_type].value_counts().reset_index()
    fleet_counts.columns = [col_fleet_type, "alerts"]
    fig_fleet = px.bar(
        fleet_counts,
        x=col_fleet_type, y="alerts",
        title="Fatigue Alerts by Fleet Type (Task Complexity)"
    )
    st.plotly_chart(fig_fleet, width="stretch")

# 3. Speed vs Hour Analysis (Environmental Factors & Workload)
if col_speed and "hour" in df.columns:
    # Remove rows with NaN speed values for this analysis
    speed_df = df.dropna(subset=[col_speed])
    if not speed_df.empty:
        fig_speed_hour = px.scatter(
            speed_df,
            x="hour", y=col_speed,
            title="Speed vs Hour of Day (Fatigue Events) - Environmental Factor",
            hover_data=[col_operator, col_asset]
        )
        st.plotly_chart(fig_speed_hour, width="stretch")

# 4. Duration vs Hour Analysis (Physiological Response)
if "duration_sec" in df.columns and "hour" in df.columns:
    fig_duration_hour = px.scatter(
        df,
        x="hour", y="duration_sec",
        title="Fatigue Event Duration vs Hour of Day (Physiological Response)",
        hover_data=[col_operator, col_asset]
    )
    st.plotly_chart(fig_duration_hour, width="stretch")

# 5. Operator vs Shift Analysis (Shift Pattern Risk)
if col_operator and col_shift:
    op_shift_counts = df.groupby([col_operator, col_shift]).size().reset_index(name="alerts")
    fig_op_shift = px.bar(
        op_shift_counts,
        x=col_operator, y="alerts", color=col_shift,
        title="Operator Fatigue Distribution by Shift (Shift Pattern Risk)"
    )
    st.plotly_chart(fig_op_shift, width="stretch")

# 6. Weekly Trend Analysis (Recovery Pattern) - With Color by Shift
if 'week' in df.columns and col_shift:
    # Create a new column for the legend
    df['shift_legend'] = df[col_shift].apply(lambda x: f"Shift {x}")
    
    # Group by week and shift
    weekly_shift_trend = df.groupby(['week', 'shift_legend']).size().reset_index(name='alerts')
    
    fig_weekly = px.line(
        weekly_shift_trend,
        x='week', y='alerts',
        color='shift_legend',
        title="Weekly Fatigue Trend by Shift (Recovery Pattern)",
        markers=True
    )
    # Customize colors for each shift
    if len(weekly_shift_trend['shift_legend'].unique()) >= 2:
        # Assign specific colors to shifts (e.g., Shift 1: blue, Shift 2: red)
        color_map = {}
        unique_shifts = sorted(weekly_shift_trend['shift_legend'].unique())
        for i, shift in enumerate(unique_shifts):
            if i == 0:
                color_map[shift] = 'blue'
            elif i == 1:
                color_map[shift] = 'red'
            else:
                color_map[shift] = f'hsl({i*60}, 70%, 50%)'  # Generate different colors for more than 2 shifts
        
        fig_weekly.update_traces(marker=dict(size=8))
        fig_weekly.update_layout(
            legend_title_text="Shift",
            legend=dict(
                orientation="h",
                yanchor="bottom",
                y=1.02,
                xanchor="right",
                x=1
            )
        )
        # Apply custom colors
        for trace in fig_weekly.data:
            if trace.name in color_map:
                trace.line.color = color_map[trace.name]
                trace.marker.color = color_map[trace.name]
    
    st.plotly_chart(fig_weekly, width="stretch")

# 7. Speed Distribution Analysis (Task Complexity)
if col_speed:
    speed_df_clean = df.dropna(subset=[col_speed])
    if not speed_df_clean.empty:
        fig_speed_dist = px.histogram(
            speed_df_clean,
            x=col_speed,
            title="Speed Distribution (Task Complexity Indicator)",
            nbins=20
        )
        st.plotly_chart(fig_speed_dist, width="stretch")


# =================== INSIGHTS BY ADVANCED ANALYTICS =====================
st.subheader("Insights by Advanced Analytics")

# 1. Critical Hour Analysis (2-5 AM)
critical_hours = [2, 3, 4, 5]
critical_alerts = df[df['hour'].isin(critical_hours)]
critical_pct = (len(critical_alerts) / len(df)) * 100 if len(df) > 0 else 0

st.markdown(f"Critical Hour Risk (2-5 AM)")
# Use conditional formatting for background color
bg_color = "#ffcccc" if critical_pct > 50 else "#ffebcc" if critical_pct > 25 else "#ffffcc" if critical_pct > 10 else "#e6ffe6"
st.markdown(f'<div style="background-color: {bg_color}; padding: 10px; border-radius: 5px;">Critical Hour Alerts: {len(critical_alerts)} ({critical_pct:.1f}% of total alerts)</div>', unsafe_allow_html=True)
if critical_pct > 10:  # If more than 10% of alerts happen in critical hours
    st.warning(f"High risk: {critical_pct:.1f}% of fatigue alerts occur during critical hours (2-5 AM). This is a known circadian dip period.")
else:
    st.info(f"{critical_pct:.1f}% of alerts occur during critical hours. This is within acceptable range.")

# 2. High-Speed Fatigue Analysis (Environmental Risk)
if col_speed:
    high_speed_threshold = df[col_speed].quantile(0.75)  # Top 25% of speeds
    high_speed_fatigue = df[df[col_speed] >= high_speed_threshold]
    high_speed_pct = (len(high_speed_fatigue) / len(df)) * 100 if len(df) > 0 else 0
    
    st.markdown(f"High-Speed Fatigue Risk (Speed > {high_speed_threshold:.0f} km/h)")
    st.metric("High-Speed Fatigue Events", f"{len(high_speed_fatigue)}", f"{high_speed_pct:.1f}% of total alerts")
    if high_speed_pct > 20:  # If more than 20% of alerts happen at high speed
        st.warning(f"High risk: {high_speed_pct:.1f}% of fatigue alerts occur at high speeds. This increases accident severity potential.")
    else:
        st.info(f"{high_speed_pct:.1f}% of alerts occur at high speeds. This is within acceptable range.")

# 3. Shift Pattern Analysis
if col_shift:
    shift_counts = df[col_shift].value_counts()
    shift_alerts_by_hour = df.groupby([col_shift, 'hour']).size().reset_index(name='alerts')
    
    st.markdown(f"Shift Pattern Risk")
    for shift_val in shift_counts.index:
        shift_pct = (shift_counts[shift_val] / len(df)) * 100
        st.metric(f"Shift {shift_val} Alerts", f"{shift_counts[shift_val]}", f"{shift_pct:.1f}% of total alerts")
        if shift_pct > 50:  # If one shift has more than 50% of alerts
            st.warning(f"Shift {shift_val} has disproportionately high alerts ({shift_pct:.1f}%). Review shift scheduling and workload.")
        else:
            st.info(f"Shift {shift_val} alert distribution is acceptable ({shift_pct:.1f}%).")

# 4. Operator Risk Profiling
if col_operator:
    operator_alerts = df[col_operator].value_counts()
    top_risk_operators = operator_alerts.head(5)  # Top 5 operators by alerts
    
    st.markdown(f"High-Risk Operator Identification")
    for op_name, count in top_risk_operators.items():
        op_pct = (count / len(df)) * 100
        st.metric(f"Operator: {op_name}", f"{count} alerts", f"{op_pct:.1f}% of total alerts")
        if op_pct > 5:  # If an operator has more than 5% of all alerts
            st.warning(f"Operator {op_name} has high fatigue risk ({op_pct:.1f}% of alerts). Consider coaching or rest plan.")
        else:
            st.info(f"Operator {op_name} fatigue risk is within acceptable range ({op_pct:.1f}%).")


# =================== FATIGUE RISK MATRIX =====================
# Moved to sidebar
with st.sidebar:
    st.subheader("Fatigue Risk Matrix")
    
    risk_matrix_data = [
        ["High fatigue + high-speed haul road", "Potential fatality", "Critical"],
        ["Moderate fatigue + decline haul road", "Serious injury", "High"],
        ["High fatigue + low-risk task", "Minor injury", "Medium"],
        ["Low fatigue + non-hazard task", "No injury", "Low"]
    ]
    
    risk_df = pd.DataFrame(risk_matrix_data, columns=["Likelihood (Fatigue Level)", "Severity (Hazard Impact)", "Risk Tier"])
    
    # Display risk matrix as a styled table
    html_string = '<table class="risk-matrix"><thead><tr><th>Likelihood (Fatigue Level)</th><th>Severity (Hazard Impact)</th><th>Risk Tier</th></tr></thead><tbody>'
    for _, row in risk_df.iterrows():
        risk_class = row["Risk Tier"].lower()
        html_string += f'<tr class="{risk_class}"><td>{row["Likelihood (Fatigue Level)"]}</td><td>{row["Severity (Hazard Impact)"]}</td><td>{row["Risk Tier"]}</td></tr>'
    html_string += '</tbody></table>'
    
    st.markdown(html_string, unsafe_allow_html=True)


# =================== AI INSIGHT ENGINE =====================
st.subheader("Automated Insight Summary")

# Create a more elegant summary
insights = []

# Peak hour
if "hour" in df.columns and not df.empty:
    peak_hour = df["hour"].value_counts().idxmax()
    critical_hours = [2, 3, 4, 5]
    if peak_hour in critical_hours:
        insights.append(f"⚠️ Most fatigue risk occurs at **{peak_hour}:00** — during critical circadian low period (2-5 AM). Consider enhanced monitoring.")
    else:
        insights.append(f"Most fatigue risk occurs at **{peak_hour}:00** — likely due to circadian drop.")

# Risk shift
if col_shift and not df.empty:
    worst_shift = df[col_shift].value_counts().idxmax()
    insights.append(f"👷 Highest fatigue recorded in **Shift {worst_shift}** — review scheduling & workload.")

# Worst operator
if col_operator and not df.empty:
    worst_operator = df[col_operator].value_counts().idxmax()
    insights.append(f"⚠️ Operator at highest risk: **{worst_operator}** — suggested coaching or rest plan.")

# Duration risk
if "duration_sec" in df.columns and not df.empty:
    avg_duration = df["duration_sec"].mean()
    if not pd.isna(avg_duration) and avg_duration > 10:
        insights.append("⏳ Long fatigue event duration suggests slow response — improve alerting training.")

# Critical hour insight
if "hour" in df.columns and not df.empty:
    critical_alerts = df[df['hour'].isin([2, 3, 4, 5])]
    if len(critical_alerts) > 0:
        critical_pct = (len(critical_alerts) / len(df)) * 100
        if critical_pct > 15:
            insights.append(f"🌙 **CRITICAL HOUR RISK**: {critical_pct:.1f}% of alerts occur during circadian low (2-5 AM). Consider enhanced monitoring during this period.")

# High-speed insight
if col_speed and not df.empty:
    high_speed_fatigue = df[df[col_speed] >= df[col_speed].quantile(0.75)] if not df[col_speed].dropna().empty else pd.DataFrame()
    if len(high_speed_fatigue) > 0:
        high_speed_pct = (len(high_speed_fatigue) / len(df)) * 100
        if high_speed_pct > 20:
            insights.append(f"🚀 **HIGH-SPEED RISK**: {high_speed_pct:.1f}% of fatigue events occur at high speeds, increasing accident severity potential.")

# Output insights in an elegant format
for i in insights:
    st.markdown(f"- {i}")


# ================= FOOTER ===========================
st.markdown("---")
st.markdown('<div class="footer">MineVision AI - Transforming Mining Safety with Intelligent Analytics | Contact: sales@minevision-ai.com</div>', unsafe_allow_html=True)