Machine Learning: Trend Pulse

//@version=5
indicator("Machine Learning: Trend Pulse", overlay=false)

// Input Setup
winSpan = input(70, "Window Size", tooltip="Length of the window for comparing historical price changes.")
k_neighbors = input(30, "Count of k-nearest Neighbors", tooltip="Number of nearest neighbors to consider for estimating the price.")
sampleSpeed = input(40, "Sampling Rate", tooltip="The step size for sampling historical data points.")
histCap = input(100, "Historical Cap", tooltip="The maximum number of past bars to consider for estimation.")

// Calculate mean and standard deviation of close prices
avgClose = ta.sma(close, histCap)
stdDevClose = ta.stdev(close, histCap)

// Feature extraction: get the last 'winSpan' price changes
deltaPrice(idx) =>
    if idx < winSpan
        0.0
    else
        close[idx] - close[idx - winSpan]

// Calculate distance between two windows, using Euclidean distance to measure the similarity between two windows of price changes
// Lorentzian Distance, Manhattan Distance, Chebyshev Distance - these are all distances tested and make miniature changes, no point to add them here as options to choose.
findDistance(i1, i2) =>
    total = 0.0
    for i = 0 to winSpan - 1
        total := total + math.pow(deltaPrice(i1 - i) - deltaPrice(i2 - i), 2)
    math.sqrt(total)

// k-NN forecast by sampleSpeed
estimatePrice(id) =>
    distArray = array.new_float(0)
    for i = winSpan to math.min(id - 1, histCap) by sampleSpeed
        array.push(distArray, findDistance(id, i))

    estimateSum = 0.0
    minDist = 99999999.0
    minIndex = 0
    for j = 1 to k_neighbors
        minDist := 99999999.0
        minIndex := 0
        for m = 0 to array.size(distArray) - 1
            if array.get(distArray, m) < minDist
                minDist := array.get(distArray, m)
                minIndex := m
        estimateSum := estimateSum + close[minIndex + 1]
        array.set(distArray, minIndex, 99999999.0)

    // Scale the forecast using z-score
    (estimateSum / k_neighbors - avgClose) / stdDevClose * 10 + 50

// Calculate estimate and plot
finalEstimate = estimatePrice(bar_index)

// Plots
emaplot = plot(finalEstimate, title='Predicted Price', color = color.new(#7c7677, 80), linewidth = 1)

// Plots an invisible line at the level of 50 for gradient fills later.
midLinePlot = plot(50, color = na, editable = false, display = display.none)

/// Fill the space between the prediction and the mid-line (level 50) with gradients, giving a nice visual indication of 'overbought' and 'oversold' regions.
fill(emaplot, midLinePlot, 70, 60, top_color = color.new(#2a3332, 75), bottom_color = color.new(#6ca800, 100),  title = "Overbought Gradient Fill")
fill(emaplot, midLinePlot, 50,  20,  top_color = color.new(#a83c91, 100), bottom_color =color.rgb(212, 33, 33),      title = "Oversold Gradient Fill")

// Plot horizontal lines at levels 20, 50, and 80 to clearly define the 'oversold', 'middle', and 'overbought' regions.
hline(20, color = color.new(#8b3131, 50))
hline(50, color = color.new(color.gray, 50))
hline(80, color = color.new(#2c5c2e, 50))

// Bool
showlb = input.bool(false, 'Show Direction prediction?', tooltip = 'display the predicted direction of movement')

// Label
if showlb and  bar_index == bar_index
    predictionDirection = finalEstimate > finalEstimate[1] ? "u" : "d"
    predictionColor = finalEstimate > finalEstimate[1] ? color.rgb(41, 92, 80) : color.rgb(85, 31, 54)
    label.new(x=bar_index, y=80, text="" + predictionDirection, color=predictionColor, style=label.style_label_center, size = size.tiny)

// Get plots
upper         = ta.highest(finalEstimate,6)
lower         = ta.lowest(finalEstimate,6)
plot(upper, color=color.rgb(23, 49, 33,70))
channel_upper = plot(upper, color = na, editable = false, display = display.none)
channel_lower = plot(lower, color = na, editable = false, display = display.none)
channel_mid   = plot(0, color = na, editable = false, display = display.none)

// Fill
fill(channel_mid, channel_upper, top_value =  upper, bottom_value = 0, bottom_color = na, top_color = color.new(#00c4e6, 85),title = "Channel Gradient Fill 11")