diff --git a/Model_IsolationForest.ipynb b/Model_IsolationForest.ipynb
index 4effc3db8d3bb58f36785cec96e46e2e8b702881..0018cbb78ef0c7d62cb6b1ee3015911fe40bf4b1 100644
--- a/Model_IsolationForest.ipynb
+++ b/Model_IsolationForest.ipynb
@@ -1,5 +1,23 @@
 {
  "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Modelling with use of Isolation Forest Algorithm\n",
+    "##### created by: Szymon Pawłowski\n",
+    "\n",
+    "The main purpose of this project is to find products which are put in a wrong place within stylecolors. This problem is an anomaly detection problem. For the final results to make comparison between models possible we hold to the stylecolors RH797-81X, SL171-99X, RV462-99X. We expect the outliers for the first two stylecolros (RH797-81X and SL171-99X). We have 43 measures and we divide it into 2 sets, the training set which consists of the measures up to 10:30 AM and the test set consists of the rest of the measures. We look for an outlier and set the parameters for IFA globally. \n",
+    "\n",
+    "Our data consist of PROXIMITY (strength of the RFID signal), TIMESTAMP (exact time of the signal), MEASUREMENT (number of measurement), StyleColor (stylecolor of an item), EPC (an item, RFID tag). \n",
+    "\n",
+    "At first, the IFA was used to seek for an anomaly in sum of the proximity signals per EPC per 1000 MS but this, as we may imagine, did not bring any good results. Similarly when it was used for only time passed per measurement. When we put into both informations, IFA randomly selected one of them and seek for an outlier in selected feature. This got better results than before as both informations had a chance to point out an anomaly. \n",
+    "\n",
+    "After those tests, the final approach was made. The new feature, distance to the nearest different EPC from the same StyleColor, was created. Taking an assumption that the speed of walk during measurements was constant, the distance is simply time passed from one EPC to another so the new feature calculates how much time passed between occuring signal from one EPC to another one within StyleColors. For balancing data in measurements (as we want model which detects outlier in a single measurement at final) the length of measurement needed to be equal for every measurement so we cut strictly observations above some time per measurement. Next, as the RFID guns use to have noisy signals the outliers in distances occurs also in non-anomaly EPCs that's why we set the rule that if EPC is marked as an outlier more than 50% number of tiems then it is really an anomaly and in the other case - not. \n",
+    "\n",
+    "With those assumptions and rules the results, comparing to the other models, are suprisingly good. Specially if we take into an account the fact that we have here special case of time-series (where DTW is mostly used in time-series) and on the other hand we have clusters of anomaly (when EPC is an outlier it gives a plenty of anomaly observations from signals) where DBSCAN works pretty good. The fact that the IFA takes one feature and gives approximately similar results shows the potential of this approach. "
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": 1,