fa مقایسه مدل شبکه عصبی مصنوعی با مدلهای رگرسیونی دادههای شمارشی در پیش بینی تعداد دفعات اهدای خون آمار زیستی Biostatistics پژوهشي Research <font face="Times New Roman" size="3"> </font><p class="MsoNormal" style="margin: 0cm 0cm 0pt text-align: justify unicode-bidi: embed direction: rtl text-autospace: text-justify: kashida text-kashida: 0% mso-layout-grid-align: none" dir="rtl"><font size="3"><b><span lang="FA" style="color: black font-family: "B Lotus" mso-bidi-language: FA mso-themecolor: text1">زمینه و هدف:</span></b><span lang="FA" style="color: black font-family: "B Lotus" mso-bidi-language: FA mso-themecolor: text1"> مدل­بندی یکی از روش‌های مهم برای تبیین رابطه بین متغیر پاسخ و مستقل می‌باشد. از آنجا که داده‌های مربوط به تعداد دفعات اهدای خون</span><span dir="ltr"></span><span dir="ltr"></span><span lang="FA" style="color: black mso-bidi-language: FA mso-bidi-font-family: "B Lotus" mso-themecolor: text1" dir="ltr"><span dir="ltr"></span><span dir="ltr"></span><font face="Times New Roman"> </font></span><span lang="FA" style="color: black font-family: "B Lotus" mso-bidi-language: FA mso-themecolor: text1">به صورت داده‌های شمارشی(گسسته)می‌باشد، جهت تبیین آن­ها</span><span dir="ltr"></span><span dir="ltr"></span><span lang="FA" style="color: black mso-bidi-language: FA mso-bidi-font-family: "B Lotus" mso-themecolor: text1" dir="ltr"><span dir="ltr"></span><span dir="ltr"></span><font face="Times New Roman"> </font></span><span lang="FA" style="color: black font-family: "B Lotus" mso-bidi-language: FA mso-themecolor: text1">مناسب‌تر است که از توزیع‌های متغیرهای گسسته مانند پواسن یا دوجمله‌ای منفی استفاده کرد. هدف از انجام این مطالعه تحلیل مدل‌های شمارشی به روش شبکه عصبی و مقایسه آن با روش‌های آماری کلاسیک و انتخاب بهترین روش برای پیش‌بینی تعداد دفعات اهدای خون می‌باشد. <o:p /></span></font></p><font face="Times New Roman" size="3"> </font><p class="MsoNormal" style="margin: 0cm 0cm 0pt text-align: justify unicode-bidi: embed direction: rtl text-autospace: text-justify: kashida text-kashida: 0% mso-layout-grid-align: none" dir="rtl"><font size="3"><b><span lang="FA" style="color: black font-family: "B Nazanin" mso-bidi-language: FA mso-themecolor: text1">روش کار:</span></b><span lang="FA" style="color: black font-family: "B Nazanin" mso-bidi-language: FA mso-themecolor: text1">در این<span style="mso-spacerun: yes"><font face="B Nazanin"> </font></span>مطالعه از داده‌های مربوط به اهدای خون که در پایگاه انتقال خون شهر­کرد جمع آوری شده</span><span dir="ltr"></span><span dir="ltr"></span><span lang="FA" style="color: black mso-bidi-language: FA mso-bidi-font-family: "B Nazanin" mso-themecolor: text1" dir="ltr"><span dir="ltr"></span><span dir="ltr"></span><font face="Times New Roman"> </font></span><span lang="FA" style="color: black font-family: "B Nazanin" mso-bidi-language: FA mso-themecolor: text1">است، استفاده گردید و چهار مدل رگرسیونی پواسن، دوجمله ای منفی و حالت های صفر انبوه آن ها با روش شبکه عصبی مصنوعی مورد مقایسه قرار گرفت. برای آموزش شبکه عصبی مصنوعی از الگوریتم آموزشی </span></font><span style="color: black font-size: 10pt mso-bidi-language: FA mso-bidi-font-family: "Times New Roman" mso-ascii-font-family: "Times New Roman" mso-hansi-font-family: "Times New Roman" mso-ascii-theme-font: major-bidi mso-hansi-theme-font: major-bidi mso-bidi-theme-font: major-bidi mso-themecolor: text1" dir="ltr"><font face="Times New Roman">BFGS(</font><a href="http://en.wikipedia.org/wiki/Broyden%E2%80%93Fletcher%E2%80%93Goldfarb%E2%80%93Shanno_algorithm"><span style="color: black text-decoration: none mso-bidi-language: AR-SA text-underline: none mso-themecolor: text1"><u><font face="Times New Roman" size="2">Broyden–Fletcher–Goldfarb–Shanno algorithm</font></u></span></a><font face="Times New Roman">)</font></span><span dir="rtl"></span><span dir="rtl"></span><span style="color: black font-family: "B Nazanin" mso-bidi-language: FA mso-themecolor: text1"><span dir="rtl"></span><span dir="rtl"></span><font size="3"> <span lang="FA"><font face="B Nazanin">و برای تعیین مناسب ترین مدل از معیار میانگین مربعات خطا </font></span></font></span><span dir="ltr"></span><span dir="ltr"></span><span style="color: black font-size: 10pt mso-bidi-language: FA mso-bidi-font-family: "Times New Roman" mso-ascii-font-family: "Times New Roman" mso-hansi-font-family: "Times New Roman" mso-ascii-theme-font: major-bidi mso-hansi-theme-font: major-bidi mso-bidi-theme-font: major-bidi mso-themecolor: text1" dir="ltr"><span dir="ltr"></span><span dir="ltr"></span><font face="Times New Roman">(mean-square error)</font></span><span dir="rtl"></span><span dir="rtl"></span><span lang="FA" style="color: black font-size: 10pt mso-bidi-language: FA mso-bidi-font-family: "Times New Roman" mso-ascii-font-family: "Times New Roman" mso-hansi-font-family: "Times New Roman" mso-ascii-theme-font: major-bidi mso-hansi-theme-font: major-bidi mso-bidi-theme-font: major-bidi mso-themecolor: text1"><span dir="rtl"></span><span dir="rtl"></span><font face="Times New Roman">(</font></span><font face="Times New Roman"><span style="color: black font-size: 10pt mso-bidi-language: FA mso-bidi-font-family: "Times New Roman" mso-ascii-font-family: "Times New Roman" mso-hansi-font-family: "Times New Roman" mso-ascii-theme-font: major-bidi mso-hansi-theme-font: major-bidi mso-bidi-theme-font: major-bidi mso-themecolor: text1" dir="ltr">MSE</span><span dir="rtl"></span><span dir="rtl"></span><span lang="FA" style="color: black font-size: 10pt mso-bidi-language: FA mso-bidi-font-family: "Times New Roman" mso-ascii-font-family: "Times New Roman" mso-hansi-font-family: "Times New Roman" mso-ascii-theme-font: major-bidi mso-hansi-theme-font: major-bidi mso-bidi-theme-font: major-bidi mso-themecolor: text1"><span dir="rtl"></span><span dir="rtl"></span>)</span></font><font size="3"><span lang="FA" style="color: black font-family: "B Nazanin" mso-bidi-language: FA mso-themecolor: text1"> استفاده شد. بهترین ساختار شبکه در داده های آموزش انتخاب و دقت روش شبکه عصبی با بهترین ساختار در داده های آموزش با مدل های رگرسیونی کلاسیک مورد مقایسه قرار گرفت تا بهترین روش برای پیش بینی تعداد دفعات مجدد اهدای خون انتخاب گردد.<span style="mso-spacerun: yes"><font face="B Nazanin"> </font></span></span><span style="color: black mso-bidi-language: FA mso-bidi-font-family: Lotus mso-themecolor: text1" dir="ltr"><o:p /></span></font></p><font face="Times New Roman" size="3"> </font><p class="MsoNormal" style="margin: 0cm 0cm 0pt text-align: justify unicode-bidi: embed direction: rtl text-justify: kashida text-kashida: 0%" dir="rtl"><font size="3"><b><span lang="FA" style="color: black font-family: "B Nazanin" mso-bidi-language: FA mso-themecolor: text1">یافته‌ها:</span></b><span lang="FA" style="color: black font-family: "B Lotus" mso-bidi-language: FA mso-themecolor: text1"> </span><span lang="FA" style="color: black font-family: "B Nazanin" mso-bidi-language: FA mso-themecolor: text1">میزان </span></font><font face="Times New Roman"><span style="color: black font-size: 10pt mso-bidi-language: FA mso-bidi-font-family: "Times New Roman" mso-ascii-font-family: "Times New Roman" mso-hansi-font-family: "Times New Roman" mso-ascii-theme-font: major-bidi mso-hansi-theme-font: major-bidi mso-bidi-theme-font: major-bidi mso-themecolor: text1" dir="ltr">MSE</span><span dir="rtl"></span><span dir="rtl"></span></font><span lang="FA" style="color: black font-family: "B Nazanin" mso-bidi-language: FA mso-themecolor: text1"><span dir="rtl"></span><span dir="rtl"></span><font size="3"> برای مدل های رگرسیونی پواسن، پواسن با صفر انبوه، دوجمله ای منفی و دوجمله ای منفی با صفر انبوه به ترتیب برابر با 71/2، 54/1، 94/0 و 01/1 و برای روش شبکه عصبی مصنوعی14:17:1 با تابع تبدیل تانژانت هایپربولیک هم در لایه میانی و هم در لایه خروجی این معیار 056/0 بدست آمد.<o:p /></font></span></p><font face="Times New Roman" size="3"> </font><p class="MsoNormal" style="margin: 0cm 0cm 0pt text-align: justify unicode-bidi: embed direction: rtl text-justify: kashida text-kashida: 0%" dir="rtl"><font size="3"><b><span lang="FA" style="color: black font-family: "B Nazanin" mso-bidi-language: FA mso-themecolor: text1">نتیجه‌گیری:</span></b><span lang="FA" style="color: black font-family: "B Lotus" mso-bidi-language: FA mso-themecolor: text1"> </span><span lang="FA" style="color: black font-family: "B Nazanin" mso-bidi-language: FA mso-themecolor: text1">نتایج مطالعه نشان داد که، با توجه به میزان </span></font><font face="Times New Roman"><span style="color: black font-size: 10pt mso-bidi-language: FA mso-bidi-font-family: "Times New Roman" mso-ascii-font-family: "Times New Roman" mso-hansi-font-family: "Times New Roman" mso-ascii-theme-font: major-bidi mso-hansi-theme-font: major-bidi mso-bidi-theme-font: major-bidi mso-themecolor: text1" dir="ltr">MSE</span><span dir="rtl"></span><span dir="rtl"></span></font><span lang="FA" style="color: black font-family: "B Nazanin" mso-bidi-language: FA mso-themecolor: text1"><span dir="rtl"></span><span dir="rtl"></span><font size="3"> می­توان روش شبکه عصبی مصنوعی را مناسب­ترین روش با بالاترین دقت جهت پیش­بینی تعداد دفعات اهدای مجدد خون نسبت به مدل های مورد بررسی در این پژوهش دانست. <o:p /></font></span></p><font face="Times New Roman" size="3"> </font> <font face="Times New Roman" size="3"> </font><p class="MsoNormal" style="margin: 0cm 0cm 0pt text-align: justify text-justify: kashida text-kashida: 0%"><font size="3"><font face="Times New Roman"><b><span style="color: black mso-themecolor: text1">Background</span></b><b><span style="color: black mso-bidi-font-family: "Times New Roman" mso-ascii-font-family: "Times New Roman" mso-hansi-font-family: "Times New Roman" mso-ascii-theme-font: major-bidi mso-hansi-theme-font: major-bidi mso-bidi-theme-font: major-bidi mso-themecolor: text1">:</span></b><span style="color: black mso-bidi-font-family: "Times New Roman" mso-ascii-font-family: "Times New Roman" mso-hansi-font-family: "Times New Roman" mso-ascii-theme-font: major-bidi mso-hansi-theme-font: major-bidi mso-bidi-theme-font: major-bidi mso-themecolor: text1"> Modeling is one of the most important ways for explanation of relationship between dependent and independent response. Since data, related to number of blood donations are discrete, to explain them it is better to use discrete variable distribution like Poison or Negative binomial. This research tries to analyze numerical methods by using neural network approach and compare it by classic statistical methods to choose better way to predict the number of blood donations.<o:p /></span></font></font></p><font face="Times New Roman" size="3"> </font><p class="MsoNormal" style="margin: 0cm 0cm 0pt text-align: justify text-justify: kashida text-kashida: 0%"><font size="3"><font face="Times New Roman"><b><span style="color: black mso-themecolor: text1">Methods</span></b><b><span style="color: black mso-bidi-font-family: "Times New Roman" mso-ascii-font-family: "Times New Roman" mso-hansi-font-family: "Times New Roman" mso-ascii-theme-font: major-bidi mso-hansi-theme-font: major-bidi mso-bidi-theme-font: major-bidi mso-themecolor: text1">:</span></b><span style="color: black mso-bidi-font-family: "Times New Roman" mso-ascii-font-family: "Times New Roman" mso-hansi-font-family: "Times New Roman" mso-ascii-theme-font: major-bidi mso-hansi-theme-font: major-bidi mso-bidi-theme-font: major-bidi mso-themecolor: text1"> In this study, data were collected from blood donors at the blood center of the Sharekord and then four methods were compared by neural network approach. These methods are: Poisson regression model and its zero inflated, Negative binomial models and its zero inflated.To learn neural network approach, (BFGS) Broyden–Fletcher–Goldfarb–Shanno algorithm was used. </span><span style="color: black mso-themecolor: text1">To choose the best model, mean-square error</span></font><span class="apple-converted-space"><span style="background: white color: black font-family: "Arial","sans-serif" mso-themecolor: text1"> </span></span><span style="color: black mso-themecolor: text1"><font face="Times New Roman">(MSE) was used. The best network structure in teaching data was chosen and neural network approach resolution was compared by them, to choose the best approach for prediction the number of blood donations.<o:p /></font></span></font></p><font face="Times New Roman" size="3"> </font><p class="MsoNormal" style="margin: 0cm 0cm 0pt text-align: justify text-justify: kashida text-kashida: 0%"><font size="3"><font face="Times New Roman"><b><span style="color: black mso-bidi-font-family: "Times New Roman" mso-ascii-font-family: "Times New Roman" mso-hansi-font-family: "Times New Roman" mso-ascii-theme-font: major-bidi mso-hansi-theme-font: major-bidi mso-bidi-theme-font: major-bidi mso-themecolor: text1">Results:</span></b><span style="color: black mso-bidi-font-family: "Times New Roman" mso-ascii-font-family: "Times New Roman" mso-hansi-font-family: "Times New Roman" mso-ascii-theme-font: major-bidi mso-hansi-theme-font: major-bidi mso-bidi-theme-font: major-bidi mso-themecolor: text1"> The MSE for Poisson regression model, Poisson regression with zero inflated, negative binomial and negative binomial with zero inflated are respectively 2.71, 1.54, 0.94 and 1.01. For neural network approach 14:17:1 with activation function of hyperbolic tangent in hidden layer and output layer 0.056 is achieved. <o:p /></span></font></font></p><font size="3"><font face="Times New Roman"> <b><span style="color: black font-family: "Times New Roman","serif" font-size: 12pt mso-bidi-language: AR-SA mso-fareast-font-family: "Times New Roman" mso-ansi-language: EN-US mso-fareast-language: EN-US mso-ascii-theme-font: major-bidi mso-hansi-theme-font: major-bidi mso-bidi-theme-font: major-bidi mso-themecolor: text1">Conclusion:</span></b><span style="color: black font-family: "Times New Roman","serif" font-size: 12pt mso-bidi-language: AR-SA mso-fareast-font-family: "Times New Roman" mso-ansi-language: EN-US mso-fareast-language: EN-US mso-ascii-theme-font: major-bidi mso-hansi-theme-font: major-bidi mso-bidi-theme-font: major-bidi mso-themecolor: text1"> </span><span style="color: black font-family: "Times New Roman","serif" font-size: 12pt mso-bidi-language: AR-SA mso-fareast-font-family: "Times New Roman" mso-ansi-language: EN-US mso-fareast-language: EN-US mso-themecolor: text1">The results showed that, according to amount of MSE, neural network approach is the best method with highest accuracy to predict the number of blood donations rather than other methods examined in this article</span></font></font> رگرسیون پواسن, پواسن با صفر انبوه, دوجمله ای منفی, دوجمله ای منفی با صفر انبوه, شبکه عصبی مصنوعی Poisson regression, Zero inflated Poisson, Negative binomial, Zero inflated negative binomial, Artificial neural network 63 70 http://rjms.iums.ac.ir/browse.php?a_code=A-10-2011-2&slc_lang=fa&sid=1 shima Haghani شیما حقانی shima_haghani@yahoo.com 3900319475328460029310 3900319475328460029310 No Shahrekord University of Medical Sciences, Shahrekord, Iran دانشگاه علوم پزشکی شهر‌کرد، شهرکرد، ایران Morteza Sedahi مرتضی سدهی sedehi56@gmail.com 3900319475328460029311 3900319475328460029311 Yes Shahrekord University of Medical Sciences, Shahrekord, Iran دانشگاه علوم پزشکی شهر‌کرد، شهرکرد، ایران Soleiman Kheiri سلیمان خیری kheiri@hbi.ir 3900319475328460029312 3900319475328460029312 No Shahrekord University of Medical Sciences, Shahrekord, Iran دانشگاه علوم پزشکی شهر‌کرد، شهرکرد، ایران